Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 514 https://internationalpubls.com A Novel Approach to Confidentiality Preservation in Big Data Using Distinct Contextual Sensitivity Hashing Dr Gowthami V1, Dr P V Kumaraguru2, S. Mohammed Nawaz Basha3, Afsal Basha V A4 1Associate Professor, School of Sciences and Computer Studies, CMR University, Chagalahatti, Bengaluru, Karnataka, India. E-mail: v.gowthami@cmr.edu.in 2COE & Associate Professor, Department of MCA, Guru Nanak College (Autonomous), Velachery, Chennai, Tamil Nadu, India. E-mail: pvkumaraguru@gurunanakcollege.edu.in 3Assistant Professor, Department of Computer Science, Islamiah College (Autonomous), Vaniyambadi,Tamil Nadu, India. E-mail: nbasha19977@gmail.com 4Assistant Professor, Department of Computer Science, Islamiah College (Autonomous), Vaniyambadi, Tamil Nadu, India. E-mail: vaafsalbasha@gmail.com Article History: Received: 05-08-2024 Revised: 14-09-2024 Accepted: 21-09-2024 Abstract: Despite the rapid development of technological innovations (IT), hypersensitive large-scale data assembling and efficiency have gotten better. To preserve the confidentiality of patients in the field of health care, it is necessary to reduce redundant confidential data whereas implementing hazardous big data sets which need to be discovered with the goal to gather appropriate data. Over the past decade, an array of preserving confidentiality approaches has been implemented employed by employing the quasi-identifier (QI) with application which includes healthcare services. Nevertheless, because of the enormous quantity of the majority of databases, protecting confidentiality across near-identifiers remains challenging in situations of enormous amounts of data. Because of datasets evolving constantly, traditional methods experience more time utilisation and reduced knowledge utility. In this paper, researchers present an advantageous Distinct Contextual Highly sensitive with Hellinger Convolutional Learner (DCS-HCL) technique that preserves anonymity yet optimising the value of information across enormous medical databases. In the beginning a Distinct Impact Contextual Delicate Hashing, or framework is generated using the input that has been provided Massive Dataset, and this model examines each of the distinct and affect variables prior implementing the results to Contextual Sensitivity Hashing. This serves as a basis enabling the development of highly computationally effective anonymous information through correlating associated QI- classes. For the purpose of preserve the confidential nature associated with personal data that is unstructured, an Hellinger Convolutional Neuro Security Conservation algorithm has been provided. This is accomplished through modifying CNN's algorithms strength and biases simultaneously processing QI-class data to maximize correctness and reduce data loss. The assessment's outcomes indicate that the approach we propose surpasses conventional approaches when it comes to of execution time, resource utility, loss of information, and correctness towards maintaining confidentiality using large-volume unorganized information sets. Keywords: Information technology, Context Sensitive Hashing, Quasi-Identifier, Hellinger, Impact, Big Data, Distinctive, and Convolutional Neural Network. mailto:v.gowthami@cmr.edu.in mailto:pvkumaraguru@gurunanakcollege.edu.in mailto:nbasha19977@gmail.com mailto:vaafsalbasha@gmail.com Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 515 https://internationalpubls.com 1. Introduction Because of increasing quantities of records that have been released, especially involves private, sensitive data regarding particular people or business organisations, protection of privacy hurdles have begun to arise. Numerous ways to reduce risks of sharing information are being developed to deal with these difficulties. Preserving sensitive information using a quasi-identifier serves as one of those approaches. Using the objective to preserve the confidentiality of these information sets, The equivalent Categories using Neural Filters (ENCC) [1] proposed an improved l-divers approach employing anatomical as a substitute for suppressing. In addition, the Cuckoo filters was employed to perform approximations set-membership analyses with the goal to enhance the speed at the rate that information was handled. While l-divers technique was employed, it was found out its execution duration were significantly less compared compared to traditional re-anonymization approaches. Furthermore, this filtering approach guaranteed the privacy of data that were continually developing frequently. Despite inadequate information-anonymization simulations, information utility has not been emphasised although protecting confidentiality as well as reducing the execution time. The present research presents a unique contextual sensitive hash algorithm so, with minimal intervention as well as increasing content significance, emerges towards computationally advantageous quasi- identifiers. Within [2], a completely novel protection framework featuring incorporated the anonymization and restoration has been laid away. This is the case because numerous investigators maintain an overwhelming belief because sensitive characteristics like pseudo-identifiers (QIDs) are capable of being separated. In a consequence, a highly sensitive QID, which stands utilising t-closeness along with l-divers was developed. Using experiments, it came to the conclusion whether the anonymization along with restoration over a particular timescale was able to be performed without preserving excellent data integrity. Adequate quality of data was preserved throughout the stipulated time frame, even though consistency along with data erosion were not specifically emphasized. As a way to deal with this problem, a Hellinger Convolutional Neural Privacy Preservation simulation is suggested throughout this study to protect private data as well as correctness through the use of revised mass as well as bias compared to the use of convolutional neural learning whereas additionally taking consideration of a distance with Hellinger. 1.1 Contributions The following is a summary of this paper's significant contributions to the literature: • This is the objective for the approach referred to as Distinct Contextual Sensitive and Hellinger Convolutional Learning (DCS-HCL) to ensure the confidentiality of enormous health care data with additionally optimising the value of data as well as reducing data loss. • Employing Distinct Impact Contextual Sensitive Hashing, researchers developed an approach with sensitive hash the fact that extends above predetermined limits along with emphasises on the time of execution, improving the usefulness of the information being analysed. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 516 https://internationalpubls.com • To improve reliability and reduce the loss of data, we additionally utilised a Hellinger Convolutional Neurological Privacy Preservation approach, a novel privacy-preserving algorithm for identifying quasi-identifiers. • Researchers compared our confidentiality methods of preservation with those that were employed in the past. In accordance with the results of the experiment, our technique scored significantly better with respect to of computation time, the precision, along with loss rate. 1.2 Organization structure The conceptual framework of this piece of writing essentially follows. The second section analyses the way enormous data security preservation techniques have advanced. Distinct Contextual Sensitive and Hellinger Convolutional Learner (DCS-HCL), the approach suggested, will be explained more thoroughly in the third section. In the fourth section, the experimental findings relating to the recommended method when compared to various popular confidentiality safeguard approaches were presented. In the end, the fifth section provides a conclusion. 2. Related works The confidential transmission of sensitive information through channels on the internet, encompassing many sectors covering health care records, surveillance footage, Web trafficking, as well as other areas, has increasingly been threatened in the past few years with the growing issue of web spoofing. In a consequence, maintaining anonymity has grown very challenging, resulting in data to get delivered erroneously. [3] Investigated at an international study on big data security conservation. Digital health care information having an opportunity to enhance outcomes for patients, forecast epidemic outbreaks in advance, and avoids illnesses that can be prevented, while raising concerns over privacy and security [4]. To address the challenges associated with privacy and security of data, another novel encryption arrangement employing a honey-based algorithm for encryption has been developed in [5]. Every business or financial institution provides data that has been believed to fall under the category of highly confidential or private and that they gain through many different individuals as a result of the exchange of data across the web. Such information need to be kept secure. Comparable generalised hierarchy (IGH), an instance of quasi-identifier, has been employed to provide a comprehensive context for identical data types. In this section a perfect approach was established employing global optimised k-anonymity [6], that substantially decreased the entire converging period. A technique enabling data warehouses that protects anonymity in the circumstances of enormous amounts of data has been laid out. in [7], for instance utilising nearest similar based clustering (NSB) with from the bottom up generalisation. Both of these characteristics addressed sensitivity without consideration for sensitivities and consequently ensured the confidentiality of the the user's data. [8] investigated in an examination concerning data preserving techniques. [9] brought out an evaluation of security for privacy using sensor having limited resources. Numerous methods have already been developed for providing significant knowledge while preserving the confidentiality of data. To deal with this situation when an individual possesses Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 517 https://internationalpubls.com numerous documents, confidential solutions of maintaining confidentiality of one's identity were recommended. The eradication of feature disclosure, nevertheless, wasn't accomplished. The two anonymity models, elevated identity-reserved diversification and improved identity-reserved concealment, have been developed in [10], for instance to address this issue and thereby decreasing error. Though the severity of the error had been minimised, difficulties involving numerous important features remaining continued to exist. This vulnerable entries have been preserved mainly a consequence of the application using bucketization approaches [11] has to deal with this issue. Using multi-record information sets, an additional bidirectional customizable generalisation method has been established in [12]. The loss of information was also substantially reduced in this particular instance through affirming its quasi-identifier anonymity and ensuring diversification among equivalent subgroups. Individuals' safeguarding their privacy becomes increasingly essential in this enormous based on information Technology era, nevertheless they are still advancements accomplished through boosting its significance, performance, along with optimal utilisation. Yet every individual's degree of confidentiality might differ. In [13], a confidentiality preserving approach was established to prevent loss of information using a hash-based anomaly identification process, thereby improving security of information as well as decreasing the expense associated with information portability. In the recent years, as the prototype of medical services has transformed from therapy to safeguard, there arises a heightening interest in healthcare sector that bestow wherever, all round the clock. With the blooming of the healthcare sector, there also arises an increasing requirement for collecting enormous amount of healthcare data with the purpose of enhancing the healthcare services. Despite the data being a valuable asset, serious privacy issue is said to occur with the leakage of sensitive information. In [14], local differential privacy was applied with the objective of providing significant accuracy. Using the foundation of the technology known as blockchain, a healthcare protection of privacy system was developed in [15]. In the present instance, medical information had previously been secured with the goal to ensure granularity control over access. The capacity for those using it to deny or add specific features to optimise management of keys were an additional significant feature. In addition, manipulation was avoided with the goal to avert disagreements or conflicts involving medical conditions, ensuring security as well as confidentiality. [16] Examined a number of safety and confidentiality risk factors associated with the health care industry and provided recommendations for addressing these kinds of problems. The primary focus was intentionally established for encrypting and anonymity in consideration. The advantages and disadvantages of implementing the guidelines for anonymity and encoding were additionally addressed. [17] performed an in-depth examination emphasising on the security and confidentiality factors associated with massive amounts of data and defining between security and privacy factors in enormous amounts of data. Although several improvements, there was still unfocused data loss. Furthermore [18], a systematic approach to selecting the seed information with the goal of categorising the information using the adaptable k-anonymity approach has been laid out. The Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 518 https://internationalpubls.com quantity of data lost and the duration of effort necessary to complete processing are both significantly reduced. A different approach centred around an approximate collection of attributes was laid out in [19], that eventually produced an appropriate compromise among delicate attribute range with quasi-identifier anonymity. Further to all of the problems previously mentioned, utility of data efficiency as well as data loss continues as significant issues. In a consequence, researchers recommend a Distinct Contextual Sensitivity and Hellinger Convolutional Learner (DCS-HCL) developing model, that additionally delivers precise access control for enormous health care data but additionally guarantees information effectiveness and minimises data loss. 3. Methodology The following section provides an in-depth formulation for both the y approach and a quasi-identifier rearrangement based approach. The structure of the approach is presented in Section 3.1. For the objective of identifying the quasi-identifier using big (unstructured) information, we clarify a Distinct Impact Contextual Sensitive Hashing technique in the following section: 3.2. The third subsection elaborates on the layout as well as the creation of confidentiality protection for unstructured information according to the previously established groupings (i.e. discovered quasi- identifier) employing Quasi-Identifier Categories. An Distinct Contextual Sensitive and Hellinger Convolutional Learning (DCS-HCL) approach's schematic representation is shown in Figure 1. Figure 1 Block diagram of the DCS-HCL approach, which stands for Distinctive Context Sensitive and Hellinger Convolutional Learning. The associated representation shows the way an enormous amount of big data database containing patients with diabetes is originally provided as inputs. Employing a Distinct Impact Contextual Sensitive Hashing method, feature segmentation is initially carried out on the starting point massive data dataset. This makes it possible for the arrangement of discrete QI-classes containing unorganised information, and this in turn enables the identification of highly computationally Big Data dataset ‘𝐷’ Distinctive Impact Context Sensitive Hashing Attribute segregation Anonymized data or output Hellinger Convolutional Neural Privacy Preservation Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 519 https://internationalpubls.com effective anonymous information (which are additionally referred to as quasi properties or pseudo- identifiers) using enormous amounts of data. 3.1 System model 3.2 Let's get started start again. look over the enormous data sets 'DS' the fact that has been collected from 129 hospitals in the United States which treated people with diabetes between the year 2000 and the year 2008. The data set, Collection [3] contains 50 different features or traits (Attr=a_1,a_2,...,a_n) of 'n' individuals. Owing according to the feature structure, the information contained within the "C"-shaped columns can potentially be categorised into four distinct classifications: quasi properties (Q=q_1,q_2,...,q_n), external variables (E=e_1,e_2,...,e_n), parameters that are sensitive (S=s_1,s_2,...,s_n), and oblivious characteristics (NS=ns_1,ns_2,...,ns_n). 3.3 Exceptional Impact model of context-sensitive hashing Massive data is employed for determining the initial quasi attributes employing the Distinct Impact Contextual sensitive Hashing (DI-CSH) technique. By employing knowledge from the past, quasi features may reveal details concerning precise identifier. To make it possible for all of these assets to be additionally considered into into consideration while establishing confidentiality, numerous research investigators have put forward an array different methods for identifying such quasi identification documents. These approaches nevertheless come with some disadvantages among which are higher turnaround demands and reduced data significance. By identifying fundamental core quasi features having the smallest degree of computational complexities and the greatest degree of data effectiveness, the recommended DI-CSH approach resolves this challenge. The anonymity is a technique employed to transform virtual identification numbers through different kinds according to the ENCC [1] methodology, although just barely guarantees confidentiality. Identifying the optimum quasi features in enormous collections continues to be challenging however. The both sides of the hand, and possessing an excessive number of quasi attributes limits the value of data, whereas on the flip side, possessing inadequate quasi features results in violations of privacy. This distinctive Impact Context Highly sensitive Hashing algorithm's objective is to identify the most beneficial quasi features in the Big Data dataset within a minimum amount during time as well as with the smallest amount of additional complexity feasible. The result is going to enhance efficiency yet preserve privacy alongside the smallest number of attainable quasi features. The second image beneath demonstrates a case study Distinct Impact Contextual Sensitive Hashing framework structure. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 520 https://internationalpubls.com Figure 3 Specimen Context-Impacts Distinctive Impact Hashing The objective for developing a Distinct Impact Contextual Sensitive Hashing method consists with identifying the quasi-attributes requiring little computational complexity while maintaining significant data effectiveness, as demonstrated from the subsequent public image employing the input data diabetic complications data as inputs. The particular value 'DV' is immediately evaluated by considering the overall amount of varying values in columns 'TV' along with the total amount of distinctive numbers in columns 'C_i', correspondingly. That distinctive characteristic has been defined as follows. 𝐷𝑉 = ∑ 𝐷𝑉[𝐶𝑖]𝑛 𝑖=1 𝑇𝑉 (1) The consequence of number IV is subsequently evaluated considering the corresponding category EC, which includes the entire assortment of entries in columns TV, and the ith column being taken into consideration, correspondingly. 𝐼𝑉 = 1 − 𝐸𝐶(𝑇𝑉−𝐶𝑖) 𝐸𝐶(𝑇𝑉) (2) The hashing algorithm 'H(qeid)' binds associated QI-classes to the position either random map QI- classes in order to enhance the effectiveness of groupings of Quasi Indicator Communities (QI- classes), that incorporate significant although numerous and unorganised information in the form of values that are unavailable or inaccurate data. This is the case due to the contrast with constantly Original data about diabetic patient Patient’s dataset Attribute diabetic patient Outline of diabetic patient attribute in a dataset 𝑝1 𝑝𝑛 𝑞1 𝑒1 𝑞2 𝑞3 𝑞4 𝑒2 𝑠1 𝑠2 𝑞5 𝑛𝑠1 𝑞1 𝑒1 𝑞2 𝑞3 𝑞4 𝑒2 𝑠1 𝑠2 𝑞5 𝑛𝑠1 𝑄𝑢𝑎𝑠𝑖 − 𝑎𝑡𝑡𝑟𝑖𝑏𝑢𝑡𝑒 𝐸𝑥𝑡𝑒𝑟𝑛𝑎𝑙 − 𝑎𝑡𝑡𝑟𝑖𝑏𝑢𝑡𝑒 𝑆𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑒 𝑎𝑡𝑡𝑟𝑖𝑏𝑢𝑡𝑒 − 𝑖𝑑𝑒𝑛𝑡𝑖𝑓𝑖𝑒𝑟 𝑁𝑜𝑛 − 𝑠𝑒𝑛𝑠𝑖𝑡𝑖𝑣𝑒 𝑎𝑡𝑡𝑟𝑖𝑏𝑢𝑡𝑒 𝑝1 𝑞1 𝑞2 … .. 𝑞𝑛 𝑒1 𝑒2 … .. 𝑒𝑛 𝑠1 𝑠2 … .. 𝑠𝑛 𝑛𝑠1 𝑛𝑠2 … .. 𝑛𝑠𝑛 𝑝𝑛 𝑞1 𝑞2 … .. 𝑞𝑛 𝑒1 𝑒2 … .. 𝑒𝑛 𝑠1 𝑠2 … .. 𝑠𝑛 𝑛𝑠1 𝑛𝑠2 … .. 𝑛𝑠𝑛 … … … … … … … … … Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 521 https://internationalpubls.com evolving datasets [1] propagated on an unpredictable manner, splitting QI-classes and preserving anonymity is going to be easier because QI-classes were closest to each other in terms to their quasi- identifiers. In this case, that we integrate Contextual Hashes to calculate every individual effect value generated for the total quantity of distinct numbers within the respective column associated with every QI- class. To add contextual hashing, which it must be essential to determine the distance that exists that separates the two simulated-identifiers (i.e., quasi encountering identifier), "qeidx" along with "qeidy." Have Dis(qeidx,qeidy) denote the distance that exists between the two quasi-identifiers. In addition to "qeid" x=(q_1 x, q_2 x,...,q_n x) with "qeid" y=(q_1 y, q_2 y,...,q_n y)" accordingly. The distance between the two places is subsequently calculated numerically in the illustration following. 𝑅𝑒𝑠 = 𝐷𝑖𝑠(𝑞𝑒𝑖𝑑𝑥, 𝑞𝑒𝑖𝑑𝑦) = √∑ 𝐷2(𝑞𝑖 𝑥, 𝑞𝑖 𝑦 ) (3) D(q_ix,q_iy) denotes the distance that exists connecting q_ix with q_iy from the formula (3) above. Another Contextual Hash function is an algorithm that is linked to the leading to deviation from equation (3) before for the purpose to generate homogeneous QI-classes. Considering either "d_i" with "d_j" represent two measurements, the function used for hashing is developed employing "(d_i,d_j)" given any pair of near-identifiers "qeidx,qeidy" accordance according to the pseudocode's presentation concerning the dualistic concept and likelihood distribution. Here is offered an arbitrary code that represents the unique impact of the contextual hash. Input: Big data dataset "DS," patients "P=P_1, P_2,..., P_n,"and attributes "Attr=A_1, A_2,..., A_n" Output: Quasi-identifiers that are optimized for computation Step 1: Set up and Initialize "qeidx" initially. and "qeidy." Step 2: Create classes "Cl=cl__1, cl__2, cl__3, cl__4" in column "C_i" Step 3: Begin Step 4: 'Attr=a_1,a_2,...,a_n' for each large data dataset 'DS' with 'n' characteristics and 'P' for patients Step 5: To twin quasi-identifiers, qeidx and qeidy, which stand for quasi encounter identifiers, Step 6: Utilise equation 1 to evaluate distinguishing value. Step 7: Apply equation (2) to evaluate the effect value. Step 8: Use equation (3) to calculate the distance between two quasi-identifiers. Step 9: If "Res(qeidx,qeidy)" d_j Step 10: Then, "Prob[H(qeidx)=H(qeidy)]" is used. Step 11: End if Step 12: If "Res(qeidx,qeidy)" d_j Step 13: Then, "Prob[H(qeidx)=H(qeidy)]" is used. Step 14: End if Step 15: End for Step 16: End for Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 522 https://internationalpubls.com Step 17: Bring back the quasi characteristics "p=Q=q_1, q_2,..., q_n" Step 18: End Algorithm 1 Quasi-Identifier with a Differential Impact Context Hash 3.3. Three distinct stages have been included within the Distinct Impact Contextual Hash Quasi- Identifier technique mentioned previously. The initial challenge goes to determine the distinctive and impactful value that corresponds to the overall amount of various numbers in column employing the big dataset (i.e., the kind of diabetes sample) as inputs. Comparable QI-classes have been assigned employing a context-dependent Hashing a distance algorithm based on this identified values. At last, corresponding indirect identifiers that are both efficient and computationally effective can be generated employing the concept of distance along with duality principles by translating comparable QI-classes. Through the application of this method of analysis, it can be accomplished to determine which are the most beneficial and most computationally efficient quasi-identifiers, minimising the overall amount of unfavourable characteristics which have been selected for pseudo-identifiers thus enhancing the practical application of the information being analysed. 3.4 Model for Privacy Preserving Convolutional Neural Networks by Hellinger The Distinct Impact Contextual Hashing Quasi-Identifier approach is employed in the suggested approach to extract attributes from unorganised data and to initialise the CNN algorithm arrangement shortly after extracting the computationally lightweight quasi-identifiers. Because a significant amount of data destruction is anticipated to happen when understanding the quasi-identifier then protecting them with the purpose of preserving privacy, an Hellinger Convolutional Neural Privacy Preservation approach is employed. In this work we compute the Hellinger Distance values in each equivalence class (i.e. class other than QI-classes) to quantify the distance and cautious scrutiny is paid to those QI-classes having minimum distance values. In this manner by quantifying the distance both the information loss is said to be minimized and at the same time when trained in CNN results in the improvement of accuracy. Then the learned Distinctive Impact Context Hash Quasi-Identifier is utilized to train a CNN for privacy preservation. The proposed privacy-preserving data analysis architecture is illustrated in figure 4. For the purpose of trying to measure the distance between two points, we evaluate a Hellinger Similarity ratios for every equivalency category (i.e., category apart from QI-classes) in the present investigation. We following closely examine these QI-classes exhibiting the shortest distance values. The measurement of the distance between two points has been designed to reduce the loss of information yet additionally resulting in an improvement in reliability if utilised with CNN. This developed The CNN network is subsequently employed for developing a Distinct Impact Contextual Hashing Quasi-Identifier with security of privacy. The fourth figure shows the recommended preserving confidentiality analysis of information methodology. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 523 https://internationalpubls.com Figure 4 Data Analysis Architecture of proposed privacy-preserving As demonstrated in the figure to the right, with a distinction among QI-classes as well as non-QI- classes, we presume that "X=x_1,x_2,...,x_n,x_iRm" along with "X=Q=q_1,q_2,...,q_n ", when "n" indicates the total number of specimens (i.e., beyond the quasi identification numbers acquired in QI- classes) alongside "m" In this section "W" represents to denote the weight, whereas "b" refers for a bias. The activating function is analytically mentioned in the following manner employing both of these factors. 𝐻𝑤,𝑏 = 𝐻(𝑥𝑖, 𝑊, 𝑏) = 𝑆𝐼𝐺𝑀𝑂𝐼𝐷 (𝑊𝑥𝑖 + 𝑏) (4) The standard deviation of activation can be determined with the sigmoid coefficient of the weighted average and a bias in equation (4) above. The starting point of the hypothesis is subsequently expressed using numerical methods as follows. 𝑃𝑖𝑛𝑖𝑡 = ∑ 𝐻𝐷(𝛼||𝛼𝑗)𝑙 𝑗=1 (5) In the equation mentioned above (5), both the letters "l" and "HD(.)" represent for the total amount of instances remaining in the large-scale data the collection with the execution of quasi-identifier authentication and the degree of similarity between two distributions of probability, accordingly. The mathematical description for this concept has been provided below. 𝐻2(𝑃, 𝑄) = ∫(√𝑑𝑃 − √𝑑𝑄) 2 (6) 𝑄𝐼 − 𝐶𝑙𝑎𝑠𝑠𝑒𝑠 𝑂𝑡ℎ𝑒𝑟 𝑡ℎ𝑎𝑛 𝑄𝐼 − 𝐶𝑙𝑎𝑠𝑠𝑒𝑠 𝐵𝑖𝑔 𝐷𝑎𝑡𝑎 𝐷𝑎𝑡𝑎𝑠𝑒𝑡 Hellinger Distance Activation function 𝑈𝑝𝑑𝑎𝑡𝑒 𝑊 𝑈𝑝𝑑𝑎𝑡𝑒 𝐵 𝑇𝑟𝑎𝑖𝑛𝑖𝑛𝑔 𝑤𝑖𝑡ℎ 𝐶𝑁𝑁 Privacy preserved data with maxmin principle Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 524 https://internationalpubls.com 'P' and 'Q' represent the two separate probability measurements (i.e., quasi experience identities as well as non-quasi encounter IDs) that are both continuously having the respect to the final measurement that has an unique likelihood value corresponding to the fact that additionally 'P' and 'Q' were continuously given the calculation (6) above. The acquired Distinct Impact cost function is subsequently expressed numerically as demonstrated below. 𝐶𝐶𝐼(𝑊, 𝑏) = [ 1 𝑛 ∑ ( 1 2 (𝐻𝑊,𝑏(𝑥𝑖) − (𝑦𝑖)) 2 ) + 𝐻2(𝑃, 𝑄)𝑛 𝑖=1 ] (7) The value of the cost function "C_CI" was calculated using equation (7) above employing the vector that is input "x_i," the resulting activation coefficient "H_(W,b)," associated Hellinger a distance "H2 (P,Q)," with the associated cost function. Model amended and constructed outcomes for both parameters 'W_ij' and 'b_i' are displayed below. 𝑊𝑖𝑗 = 𝑊𝑖𝑗(𝑙) − 𝐿𝑅 𝜕 𝜕𝑊𝑖𝑗(𝑙) (8) 𝑏𝑖 = 𝑏𝑖(𝑙) − 𝐿𝑅 𝜕 𝜕𝑏𝑖(𝑙) (9) Ultimately, the power source Distinct Effect costing function's average square error is evaluated as demonstrated below. 𝐶(𝑊, 𝑏) = [ 1 𝑛 ∑ ( 1 2 (𝐻𝑤,𝑏(𝑥𝑖) − (𝑦𝑖) 2))𝑛 𝑖=1 + 𝑦 2 ∑ ∑ ∑ 𝑊𝑖𝑗(𝑙)𝑛 𝑙=1 𝑛 𝑗=2 𝑛 𝑖=1 ] (10) Furthermore, through employing the initially collected and transformed information within the Big Data datasets for training The CNN network for classifying and ensure preserving the confidential nature of the data, the efficiency of the recommended Distinct Impact cost function is shown. Here you will find Hellinger Convolutional Neural Privacy Preservation's pseudo code implementation. Input: 'X=x_1, x_2,..., x_n,x_iRm' is the input vector. Output: Identifiers with little privacy loss and accuracy Step 1: Set up Weight "W" and Bias "B" Step 2: Begin Step 3: For each 'X' input vector Step 4: Equation (4) may be used to mathematically define the activation function. Step 5: Using equation (5), determine the origination hypothesis. Step 6: Equation (6) should be used to compare two probability distributions. Step 7: Equation (7) is used to mathematically define the learnt Distinctive Impact cost function. Step 8: Using equations (8) and (9), update the parameters weight and bias. Step 9: Equation (10) should be used to calculate the Distinctive Impact cost function's mean square error. Step 10: Return (identifiers protected by privacy) Step 11: End for Step 12: End Algorithm 2 Hellinger Convolutional Neural Privacy Preservation Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 525 https://internationalpubls.com The three stages are carried out as stated by the Hellinger Convolutional Neural Privacy Preservation method above. The approach proceeds to develop the maxmin approach (i.e., optimise correctness and reduce data loss), assure the preservation of privacy for data that is not structured, and safely protect highly confidential information commencing with the set of non-QI-classes provided as input. Throughout our investigation, approach is achieved by initially creating a function to activate by hashing QI-classes, subsequently applying Hellinger proximity to minimise the data loss using low connection values. Following that, employing modified the CNN network, it is provided as input for acquiring knowledge, modifying weighting and discrimination employing the Distinct Impact cost function. In such a manner, it is stated that correctness as data loss have been improved and preserving the anonymity of essential data that is unorganized. 4. Experiments and discussion The effectiveness provided by the Distinct Contextual Sensitive and Hellinger Convolutional Learner (DCS-HCL) technique for the preserving the privacy for confidential unprocessed enormous medical information with quasi-identifier is being examined in the following section thorough an in-depth investigation of the experimental results. On the basis of current cutting-edge methods identified in the published literature, an evaluation regarding the security conservation for big medical records employing quasi identifiers is performed in regards to computation time, accuracy, as well as data loss alongside considering the total number of individuals. The two privacy methods for data preservation are additionally compared: incorporated the anonymization as well as reconstructed [2] as well as The equivalent Classes using the Cuckoo Filter (ENCC) [1]. 4.1 Dataset description Throughout the studies we conduct, the Diabetes Statistics from 130-US institutions during seasons 1999-2008 Information Collection [20] is utilised. The information set contains 50 parameters which indicate individual and healthcare provider outcomes throughout an interval of ten decades of medical care provided by 130 hospitals in the United States together with delivery networks. The individual quantity, age, race, gender, gender identity, enrollment type, the length of endure in the hospital, the health care field associated with the accepting medical professional, the number of tests conducted by laboratories run, the haemoglobin A test outcome, being diagnosed, the amount of medicines, diabetics medications, numbers of outpatient services inpatient treatment, and emergency room appointments in the year preceding to the being hospitalised are only a few of the parameters that might be obtained from this data set. the Python experiments for preserving privacy are carried through with the assistance provided dataset. Performance metrics The outcomes of the factors considered to be taken for maintaining privacy are outlined in this subsection. In terms of the total quantity of individuals taken into consideration during the training, they undergo execution time, correctness, alongside data loss. 4.1.1 Run time evaluation The total duration of period need ought to be reduced to a minimal since patients will be accountable for providing enormous healthcare data and it's going to become available to the general public. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 526 https://internationalpubls.com Alternatively, data could get lost or anonymity could be compromised. nevertheless has been suggested because protecting the confidentiality of enormous healthcare data requires an enormous amount of time. The numerical representation corresponding to the execution time required can be seen below. 𝑅𝑇 = ∑ 𝑃𝑖 ∗ 𝑇𝑖𝑚𝑒 [𝑃𝑃]𝑛 𝑖=1 (11) The period of time 'RT' essential in order to safeguard the anonymity of enormous health care information employing quasi identification numbers can be determined from formula (11), wherein 'RT' varies depending on the total number of individuals considered into consideration throughout simulated 'P_i' and the amount of time used to preserve anonymity "Time [PP]'. Milliseconds of (ms) are employed because a measure of measurements. Accuracy evaluation The accurateness of quasi-identifiers needs to be preserved, and this is an essential concern for ensuring the confidentiality of enormous medical records. In another word, the degree of precision in this particular instance pertains to the extent to which personally identifiable information is maintained safe throughout the preserving confidentiality employing of quasi-identifiers method. The equation expressing the value of an accurate measurement is given below. 𝐴 = ∑ 𝑃𝐴𝑃 𝑃𝑖 𝑛 𝑖=1 ∗ 100 (12) Using the equation mentioned above (12), the degree of accuracy 'A' can be determined by considering the total number of patients brought into consideration for the model 'P_i' with the level of accuracy of the patient's dataset maintained 'P_AP'. subsequently is determined in percent (%) terms. 4.1.2 Information loss evaluation Ultimately, significant details have been stated that may have been misplaced in the endeavour preserving the anonymity of large-scale health care data. In the sense that a larger quantity of data is preserved, the risk of loss ought to be minimised. The resulting table contains an equation-based assessment of the data loss. 𝐼𝐿 = ∑ 𝑃𝑑𝑐 𝑃𝑖 𝑛 𝑖=1 ∗ 100 (13) The degree of data loss (IL) can be calculated by the quantity of individuals who were considered into consideration during the simulated event (P_i) with the number of records pertaining to patients that were compromised and breached (P_dc) throughout the protection of privacy (13). It is stated in percentage (%) terms. Discussion The significance associated with the recommended approach Distinct Contextual Sensitive and Hellinger Convolutional Learning (DCS-HCL) is examined in the following subsection using the data from the diabetes 130-US healthcare facilities dataset. The effectiveness of private data the preservation in the approach we developed was subsequently evaluated against the fact that provided Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 527 https://internationalpubls.com by two different approaches, including The equivalent Classes alongside Cuckoo Filter (ENCC) [1] and its combined anonymity and the reconstruction [2], employing all three frequently employed assessment metrics: execution time, preciseness, alongside data loss. 4.1.3 Performance measure of run time The actual run-time evaluation of efficiency must be performed first. Considering 10 different combinations using "P," Table 1 compared the recommended DCS-HCL's execution times with those of the two currently employed techniques, ENCC [1] as well as combined anonymity and reconstructing [2]. Those represent the median findings. In order to accommodate the increase in records and the corresponding equivalent quasi-identifiers, an increase in "P" quantity expands the duration of run times for all three approaches. Since the recommended approach only selects the optimal identifier to serve as quasi-identifiers, the execution times associated with the suggested approach can frequently be less compared to that of the existing methods [1] and [2]. Table 1 DCS-HCL, ENCC, and integrated anonymization and reconstruction analysis runtime findings [1] Number of patients Run time (ms) DCS-HCL ENCC Integrated anonymization and reconstruction 500 42.5 57.5 72.5 1000 75.35 105.35 125.35 1500 90.25 125.45 140.55 2000 105.35 140.55 175.55 2500 125.45 195.35 225.35 3000 140.55 215.25 255.85 3500 175.35 225.35 315.55 4000 190.15 240.55 335.25 4500 200.35 280.15 350.55 5000 225.55 315.55 385.55 Figure 5 Runtime graphically Representation Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 528 https://internationalpubls.com On the Diabetes Association's 130-US healthcare institutions dataset, the following figure demonstrates the execution times associated with the hypothesised DCS-HCL approach and compare the results with both of the previously common used methods [1] and [2]. The duration of the process expands exponentially increasing the number of individuals engaged throughout privacy preservation, corresponding to this figure. The total simulation computational time over '500' individuals employing DCS-HCL has been determined to vary between '42.5ms', '57.5ms' employing [1] and '72.5ms' using [2]. The following was calculated based on experiments carried out for '500' amounts of individuals for preserving the anonymity of significant medical information employing quasi identification numbers, alongside the total amount of time present for preserving a single patient having been '0.085ms'. Based to the outcomes, DCS-HCL was demonstrated to be capable of a duration of operation which was far shorter as [1] and [2]. The adoption of the Distinct Impact Contextual Sensitive Hashing (DI-CSH) approach subsequently led to the improvements. Through the use of this method of analysis, fundamentally essential quasi features can be determined through contrasting the hash function values of associated QI-classes with those of QI-classes that have been mapping randomly. The result minimises the amount of CPU time expected for preserving the anonymity of enormous health care information with DCS-HCL to less over 28% when compared with [1] and 42% compared to [2], correspondingly. Performance measure of accuracy Consistency metric on responsiveness and the functional analysis of correctness will be looked in immediately. Likewise employing 10 different combinations of "P," findings were conducted to compare the proposed DCS-HCL's accuracy with the results of current techniques, ENCC [1] and its combined confidentiality and reconstructions [2]. At table 2, outcomes are presented. The results of this study demonstrate that the preciseness metrics for each of the three methods fall since when 'P' factor is elevated. By minimizing the loss of data by means of proximity measuring, the suggested approach often yields results that are significantly greater in precision compared with the currently employed privacy-preserving alternatives. The present confidentiality preservation methods in contrast, fail to employ the concept of proximity measurement in order to handle data loss therefore as an outcome, provide considerably lower accuracy. Table 2 Accuracy analysis findings employing integrated anonymization and reconstruction [2], DCS-HCL, and ENCC [1] Number of patients Accuracy (%) DCS-HCL ENCC Integrated anonymization and reconstruction 500 97 95 92 1000 96.35 92.15 90.25 1500 96.15 90.55 88.35 2000 96 88.35 86.15 2500 95 86.25 84.35 3000 94.35 85.15 82.15 Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 529 https://internationalpubls.com 3500 94.15 84.35 80 4000 94 82.15 78.85 4500 93.25 81.55 75.35 5000 92 80 75 Figure 6 Graphical representation of accuracy The median precision results for each of the three distinct methods, DCS-HCL, [1] and [2], are presented as Figure 6 above. The previously shown plot illustrates that because the total number of individuals receiving treatment grows performance reduces because the initial readings associated with QI have not been altered similarly. The rationale that demonstrates the suggested technique's outstanding precision beyond the methods previously proposed [1] and [2] revolves around the reality that the proposed quasi-identification method ensures a minimum amount of distance consistency. The overall precision of employing DCS-HCL has been found to have been '97%', '95%' with [1] as '92%' with [2], including '500' numbers of individuals examined for simulating to evaluate the confidential integrity of enormous healthcare data and '485' instances of patient records accurately preserved. Hellinger Convolutional Neural Privacy Preserving an algorithm adoption as yet another factor leading to that enhancement. This maxmin notion can be applied to data that is unstructured by employing that approach. For the purpose of trying to accomplish this goal, Hellinger proximity is employed upon encoding QI-classes in order to generate a mechanism for activation which optimises correctness and ensuring confidentiality. In such a manner, it is stated that employing DCS-HCL leads to an accurate confidentiality being preserved with enormous medical records that has been improved up 10% when compared with [1] and 14% compared with [2]. 4.1.4 Performance measure of information loss Lastly, this part addresses the data loss that has been suffered. The loss of data rates were evaluated and contrasted compared the final results of both of the currently employed anonymity preservation techniques [1] and [2] with the goal to demonstrate even more the effectiveness of the recommended approach. Data are shown in table 3. Considering the proposed DCS-HCL method with earlier preservation of privacy methodologies, [1] and [2], this substantially decreased the loss of data value. The recommended approach employs the Hellinger Distance approach for preserving anonymity and Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 530 https://internationalpubls.com preserves the coherence among the QI's values, thereby improving data's potential utility and reduces data loss. Table 3 Integrated anonymization and reconstruction [2], ENCC [1], and DCS-HCL [2] analysis outcomes of information loss Number of patients Information loss (%) DCS-HCL ENCC Integrated anonymization and reconstruction 500 3 5 8 1000 3.5 6.25 9.35 1500 4 6.55 10 2000 4.25 6.85 10.55 2500 4.45 7 10.85 3000 4.85 7.25 11.35 3500 5 7.45 11.85 4000 6.35 7.85 12.45 4500 8 9 14 5000 8.15 10.15 15.35 Figure 7 Graphical representation of information loss Data eliminated by employing all three different approaches to represent it graphically in the example shown above. The diagram indicates the amount of data loss increase exponentially as the total amount of individuals receiving treatment increases. The reason for this is because of the simple reason that because the array of individuals receiving treatment expands, expands the challenges associated with preserving privacy, that ultimately compromise critical unorganised information. The total amount of information loss applying DCS-HCL was determined as being '3', '5' utilising [1] with '8' utilising [2] in accordance depending on the scenarios operate in order to preserve privacy using '500' instances of individuals, '15' amount of medical records breached during the entire procedure, and the total amount of information erosion. Given what has been discovered, it might be Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 531 https://internationalpubls.com determined that DCS-HCL experiences substantially lesser data degradation over [1] and [2]. The implementation of the Hellinger Convolutional Neural Privacy Preservation algorithm is ultimately generated the enhancement in performance. This machine learning model's deployment leverages the Distinct Impact cost function to modify weights and loss, that improves the utility of the information being analysed. Due to the aforementioned function, the usefulness of data gets better, lowering the probability of losing information is a result of this. In accordance with [1] and [2], the data showed that degradation using DCS-HCL has been reduced with 31% of whom and 56%, which is significantly. 5. Conclusion In this section, has been examined the challenge of figuring out how to successfully employ quasi- identifiers form huge, complicated medical datasets while upholding standards of confidentiality and optimising the value of the data with the smallest possible amount of execution time as well as data degradation. By removing some of the most essential quasi features from a data resource, we have proposed the Distinct Impact Contextual Sensitive Hashing (DI-CSH) framework for preserving privacy. This technique allows access to just a selected portion of the important quasi features in an information asset as opposed to the entire set of documentation, when needed by earlier methods. The Hellinger Convolutional neural Privacy Preservation method will continue to preserve the information using a maxmin approach to further enhance the effectiveness of the private-preserving technology. A QI-group will consequently span quite smaller number of terminals having a lower likelihood of data. loss. The efficiency of privacy preserving on big medical data sets could potentially be significantly improved employing our method in regard to computation time, reliability, as well as information loss comparing with conventional techniques, as proven by the evaluation findings utilising the diabetes-related 130-US institutions database. References [1] Odsuren Temuujin, Jinhyun Ahn, Dong-Hyuk Im, “Efficient L-Diversity Algorithm for Preserving Privacy of Dynamically Published Datasets”, IEEE Access, Sep 2019 [Equivalence Classes with Cuckoo Filter (ENCC) [1]] [2] Yuichi Sei, Hiroshi Okumura, Takao Takenouchi, and Akihiko Ohsuga, “Anonymization of Sensitive Quasi- Identifiers for l-Diversity and t-Closeness”, IEEE Transactions on Dependable and Secure Computing, Vol. 16, No. 4, Aug 2019 [integrated anonymization and reconstruction(2)] [3] Mohammed Binjubeir, Abdulghani Ali Ahmed, Mohd Arfian Bin Ismail, Ali Safaa Sadiq, Muhammad Khurram Khan, “Comprehensive Survey on Big Data Privacy Protection”, IEEE Access, Jan 2020 [4] Karim Abouelmehdi, Abderrahim Beni‑Hessane, Hayat Khaloufi, “Big healthcare data: preserving security and privacy”, Journal of Big Data, Springer, Jul 2018 [5] Gayatri Kapil, Alka Agrawal, Abdulaziz Attaallah, Abdullah Algarni, Rajeev Kumar, Raees Ahmad Khan, “Attribute based honey encryption algorithm for securing big data: Hadoop distributed file system perspective”, PeerJ Computer Science, Feb 2020 [6] Waranya Mahanan, W. Art Chaovalitwongse, Juggapong Natwichai, “Data anonymization: a novel optimal k- anonymity algorithm for identical generalization hierarchy data in IoT”, Service Oriented Computing and Applications, Springer, Feb 2020 [7] P. Srinivasa Rao, S. Satyanarayana, “Privacy preserving data publishing based on sensitivity in context of Big Data using Hive”, Journal of Big Data, Springer, Aug 2018 [8] P. Ram Mohan Rao, S. Murali Krishna, A. P. Siva Kumar, “Privacy preservation techniques in big data analytics: a survey”, Journal of Big Data, Springer, Jul 2018 Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2 (2025) 532 https://internationalpubls.com [9] Inayat Ali, Eraj Khan, Sonia Sabir, “Privacy-preserving data aggregation in resource-constrained sensor nodes in Internet of Things: A review”, Future Computing and Informatics Journal, Elsevier, Jan 2018 [10] Jinyan Wang, Kai Du, Xudong Luo, Xianxian Li, “Two privacy-preserving approaches for data publishing with identity reservation”, Knowledge and Information Systems, Springer, Jun 2018 [11] Razaullah Khan, Xiaofeng Tao, Adeel Anjum, Haider Sajjad, Saif ur Rehman Malik, Abid Khan, Fatemeh Amiri, “Privacy Preserving for Multiple Sensitive Attributes against Fingerprint Correlation Attack Satisfying c- Diversity”, Wireless Communications and Mobile Computing, Wiley, Jan 2020 [12] Xinning Li, Zhiping Zhou, “A generalization model for multi‑record privacy preservation”, Journal of Ambient Intelligence and Humanized Computing, Springer, Sep 2019 [13] Chandramohan Dhasarathan, Vengattaraman Thirumal, Dhavachelvan Ponnurangam, “A secure data privacy preservation for on-demand cloud service”, Journal of King Saud University – Engineering Sciences, Elsevier, Dec 2015 [14] Jong Wook Kim, Beakcheo Jang, Hoon Yoo, “Privacy-preserving aggregation of personal health data streams”, PLOS ONE | https://doi.org/10.1371/journal.pone.0207639 November 29, 2018 [15] Jie Xu, Kaiping Xue, Shaohua Li, Hangyu Tian, Jianan Hong, Peilin Hong , and Nenghai Yu, “Healthchain: A Blockchain-Based Privacy Preserving Scheme for Large-Scale Health Data”, IEEE Internet of Things Journal, Vol. 6, No. 5, Oct 2019 [16] Karim Abouelmehdi, Abderrahim Beni‑Hessane and Hayat Khaloufi, “Big healthcare data: preserving security and privacy”, Journal of Big Data, Springer, Feb 2018 [17] Priyank Jain, Manasi Gyanchandani and Nilay Khare, “Big data privacy: a technological perspective and review”, Journal of Big Data, Sep 2016 [18] Karuna Arava, Sumalatha Lingamgunta, “Adaptive k-Anonymity Approach for Privacy Preserving in Cloud”, Arabian Journal for Science and Engineering, Springer, Jul 2019 [19] C. Wafo Soh, L. L. Njilla, K. K. Kwiat, C. A. Kamhoua, “Learning quasi‑identifiers for privacy‑preserving exchanges: a rough set theory approach”, Granular Computing, Springer, Aug 2018 [20] Beata Strack, Jonathan P. DeShazo, Chris Gennings, Juan L. Olmo, Sebastian Ventura, Krzysztof J. Cios, John N. Clore, “Impact of HbA1c Measurement on Hospital Readmission Rates: Analysis of 70,000 Clinical Database Patient Records”, BioMed Research International, Hindawi Publishing Corporation, Apr 2014