Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 87 https://internationalpubls.com Implementing Adaptive Neuro-Fuzzy Inference Systems (ANFIS) for Risk Assessment of Drug Interactions Ashu Nayak1, Kapesh Subhash Raghatate2 1Assistant Professor, Department of CS & IT, Kalinga University,Raipur, India. 2Research Scholar,Department of CS & IT, Kalinga University,Raipur, India. Article History: Received: 03-09-2024 Revised: 14-10-2024 Accepted: 28-10-2024 Abstract: Novel drug development is time-consuming, difficult, and often unsuccessful. This has made combining therapies increasingly common and profitable in recent years. Healthcare professionals in the pharmaceutical sector are interested in the combination, but we must instantly solve drug-drug interactions. In few cases, single-perspective DDI evaluations are insufficient. Pharmacological therapy and patient safety depend on accurate drug interaction risk assessments. This work uses an Adaptive Neuro-Fuzzy Inference System (ANFIS) to assess and predict medication interaction hazards. The proposed method uses neural networks' generative skills and fuzzy logic to construct a robust decision-making framework that can handle drug interaction scenarios and their unpredictability. Fuzzy rules provide all the necessary risk assessment components. This includes dose, patient health issues, drug interaction profiles, and pharmacological properties. Accuracy, specificity, and Mean Squared Error evaluate the model after training with known drug interactions. These metrics evaluate model performance. These measures are essential for machine learning evaluation. The ANFIS-based model outperforms previous risk assessment methods in risk classification and prediction. This study found that the ANFIS architecture may improve medication management security and prevent dangerous drug interactions. Keywords: Drug Interaction, Adaptive Neuro-Fuzzy Interference, Decision making, Mean Square Error. 1. INTRODUCTION Drug design takes time and effort. Clinical studies require many processes, from objective setting to execution. Each computational phase, from target discovery to clinical trials, can use various computational methods. Figure 1 shows all computational tools relevant across drug development phases in a flow diagram. We invented some great computational systems and methodologies for drug discovery and development. This field includes target verification, virtual screening based on docking, scoring functions, conformation sampling, molecular resemblance computing, virtual library construction, and sequence-based drug design. Fig.1: Drug Design Process Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 88 https://internationalpubls.com Life expectancy is increasing as a result of better medical treatment and living situations [20]. The rising rates of drug interactions and adverse drug events, along with the prevalence of many chronic diseases in the elderly, make individuals and healthcare systems vulnerable to the devastating effects of polypharmacy and multi-morbidity [2]. Due to the sector's complexity, combination drugs are becoming more common in pharmacotherapy. To improve patient outcomes and treatment efficacy. Novel medication production still has a high failure rate, long delays, and expensive costs. Even when certain combinations show promising benefits, harmful drug interactions must be addressed immediately to protect patients. Due to the complexity of these linkages, typical DDI evaluation methods are reductionist. Understanding the multiple elements that affect DDI is essential for management. These considerations include prescription doses, patient health, drug pharmacodynamics and pharmacokinetic characteristics [4] [14]. Due to the complexity of the scenario, risk assessment must be stricter to ensure safe and effective pharmaceutical therapy. This study applies the ANFIS to risk assessment and drug interaction prediction. This study aims to address the aforesaid issues. Fuzzy logic and neural networks make the ANFIS a smart decision- making tool that simplifies drug interactions' complexity and unpredictability [8]. Fuzzy rules allow the model to assess risk by include key medication interaction data [6]. This study found that using the ANFIS framework can improve medication management by emphasizing patient safety and reducing adverse drug interactions. Advanced computational methods in clinical practice may lessen complex pharmacological hazards as the healthcare sector rapidly changes [10] [18]. Key benefits of the article 1. This study assessed and predicted pharmaceutical interaction risks using an ANFIS. The framework uses fuzzy logic and neural networks to handle drug interaction complexity and unpredictability, making it a powerful decision-making tool. 2. After training with a dataset of known drug interactions, the model is assessed for accuracy, specificity, and MSE. 3. This study suggests that the ANFIS architecture may improve medication management security and reduce dangerous drug interactions. 2. LITERATURE SURVEY Han et al. [9] say the MCFF-MTDDI model can predict several DDI types. Initial phases were KG characteristics, pill-matching label data, and medicine chemical structure extraction. A multi-channel feature fusion module completed the package. Finally, the fully-connected neural network accurately predicted various DDIs [12]. Implementing a Gated Recurrent Unit-based multi-channel feature fusion module reduced feature redundancy. We shall discuss drug combinations in depth here. The multi-class and multi-label prediction tasks tested MCFF-MTDDI's predictions about drug interactions between known-new, new-new, and known-new pharmaceuticals using four datasets [16]. All findings showed that the MCFF-MTDDI formula worked.The Dutch Pharmacogenetics Working Group (DPWG) develops evidence-based recommendations to improve pharmacotherapy and PGx use, according to Hulshof et al. [22]. This idea adjusts the initial dosage of the anti-cancer drug irinotecan to lessen the risk of febrile neutropenia and diarrhea. Mutations in the uridine diphosphate glucuronosyl transferase 1A1 gene increase the risk of immunosuppression from irinotecan. The DPWG advises leaving intramuscular (IM) irinotecan at 70% for PM patients starting Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 89 https://internationalpubls.com treatment. To receive irinotecan, patients must undergo "important" UGT1A1 genotyping tests, according to the DPWG clinical consequence score. Tang et al. [11] invented the pluggable substructure interaction module DSIL-DDI, which stands for domain consistent substructure contact of drug-drug prediction. With this module, the source domain can acquire domain-invariant DDI representations. It studies substructure interactions. The first scenario, transduction, includes all drugs in the test set and the training set. The second scenario, induction, adds new medications to the training set. In the third and final situation, "out-of-the-box generalization," testing and training databases are different. The findings suggest that DSIL-DDI could improve OOD DDI predictions and help models generalize and comprehend data. Medical personnel can secure patient medicine delivery and decrease drug abuse side effects with DSIL-DDI. Gill et al. [19] tested Reg-ML's ability to predict drug exposure from pharmacokinetic drug-drug interactions. This study examined demographics, cytochrome P450 metabolic activity, in vitro pharmacokinetic parameters, organometallic content, and physicochemical properties. We used fivefold cross-validation to evaluate the model. A support vector regression model predicted 78% of exposure changes within two standard deviations. These findings suggest that machine learning algorithms can predict drug exposure using early drug development data.Lu et al. [13] developed a statistical methodology to examine databases of spontaneous adverse events, drug-host interactions, and host-specific risk alterations. The framework for identifying safety signals combines several methods. A regular approximation test, likelihood ratio test, and two subgroup ratio tests comprised our four-pronged method. Each test was part of the approach. We examined the FDA's Adverse Event Reporting Systems (FAERS) to investigate if gender and age affected liver event reporting for specific therapy classes. Use of data achieved this. The simulation showed that the normal approximation and likelihood ratio can reduce family-based error rates and identify host-related adverse medication events. 3. SYSTEM METHODOLOGY 3.1 Overview of Adaptive Neuro-Fuzzy Inference System (ANFIS) Figure 1 depicts the suggested DDI risk assessment and prediction model utilizing the ANFIS. This model can be organized into the following essential parts. Fig. 2: Overview of Adaptive Neuro-Fuzzy Inference System The input data includes drug interaction reports with dosage information, Age and comorbidities of patients, Drug metabolism, therapeutic class Contraindications, severity, and drug interactions. Finding relevant features reduces complexity and processing time. Normalizing data ensures neural Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 90 https://internationalpubls.com network processing compatibility.Features become hazy linguistic variables. So, the dosage can be low, medium, or high. Minor, Moderate, and Severe health issues exist. Fuzzy rules represent input variable interaction logic. The risk is high if the dosage and health concernsare severe. The Fuzzy Membership Function Membership functions (triangular) fuzzily feature for nuanced risk classifications. The ANFIS framework combines fuzzy inference with neural network learning. The neural network modifies fuzzy rules and membership functions based on training data.The training approach uses hybrid learning (gradient descent and least-squares estimation) to reduce prediction errors. Dataset- trained and verified models minimize Mean Squared Error (MSE) and maximize classification precision.The fuzzy output (e.g., High, Medium, and Low Risk) is defuzzed into a crisp value for final decision-making. The defuzzified output classifies DDI risk as Safe, Caution, or Dangerous.Calculates the percentage of high-risk interactions predicted accurately. Testing the model's negative case detection. Estimates the model's prediction accuracy by averaging the squared differences between actual and anticipated values.The system shows how fuzzy rules classify DDI risks in an interpretable decision-making framework. Pharmaceutical professionals can visualize risk levels for better analysis.The suggested approach outperforms current DDI risk assessment methods when verified against benchmarks. Clinical decision support systems (CDSS) can use the ANFIS- based model to evaluate drug interactions in real-time to improve patient safety and reduce adverse drug reactions.This structure allows the ANFIS-based system to handle complicated and dynamic drug-drug interactions, making it a reliable medication management solution. 3.2 Working of ANFIS A well-selected model, the ANFIS, can help with risk and uncertainty management. This study has two objectives. First, let's examine the current state of flow risk estimation. Second, intelligent risk assessment and management measures, or indicators, should be established to comprehend railway station real-time safety better. An artificial neural network (ANN) plus a formal reasoning system (FIS) constitute an ANN-FIS. Mixing the financial information system (FIS) with a flexible network structure created the ANFIS. The ANFIS model is gaining popularity among technical and scientific academics due to its remarkable learning and reasoning abilities. Concerns about parameter identification use ANFIS. Its hybridlearning rule uses leastsquares, gradient descent, and back- propagation to achieve this. Figure 3 shows this multilayer feed-forward network's input-output mapping using an artificial neural network and fuzzy logic. So, it's an FLS embedded in an adaptive ANN. ANFIS's capabilities allow it to build IF/THEN network implementations. Fig. 3: Working of ANFIS Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 91 https://internationalpubls.com ANFIS model for drug-drug interaction (DDI) risk assessment manages pharmaceutical interaction complexity in a hybrid framework. Combining fuzzy logic's interpretability with neural networks' learning capacities does this. The model predicts medication interactions using several essential input variables, such as dosage and therapeutic classification, demographic information like Age and pre- existing medical conditions and pharmacological characteristics like medication metabolism and contraindications. Convert these input values into fuzzy language phrases (e.g., Low, Medium, High for dosage levels) to better understand interaction possibilities. One imprecise guideline that describes these elements is that the risk is significant if the dosage is high and the patient's health issue is severe. \Since these fuzzy rules compose the fuzzy inference system, the model can uniformly and unambiguously assess risk levels. The neural network component of the ANFIS model optimizes fuzzy membership functions and rules during training. The neural network uses gradient descent and least-squares estimation to minimize prediction error and maximize risk categorization accuracy. The model adjusts fuzzy inference system settings based on training results. These changes help the model understand complex drug interactions on a large dataset of known DDIs. This training approach produces a successful adaptive system for real-time applications since it can generalize to novel and unknown drug interactions. The defuzzification module converts training outputs into precise risk scores. Thus, classifying interaction hazards as low, medium, or high is easier. Our decision-making system exceeds cutting-edge risk assessment methods in industry-standard performance parameters like Mean Squared Error (MSE), Precision, Specificity, and Sensitivity. Due to its comprehensiveness, the DDI risk assessment tool is reliable and accurate throughout drug management. It aids clinical decision-making and reduces hazardous drug interactions. Users can understand fuzzy rule-based DDI risk classifications thanks to the system's interpretable decision- making framework. Visualizing risk levels may help pharmaceutical professionals understand and analyze data. 4. EXPERIMENTAL RESULTS The proposed model utilizes the Comprehensive DDI Information (https://paperswithcode.com/dataset/ddi). Datasets that provide extensive information about pharmaceutical interactions, patient health records, and pharmacological characteristics are required by the proposed ANFIS model in order to estimate the likelihood of drug-drug interactions (DDIs). It is important for a high-quality dataset to have various variables because this will facilitate the learning and evaluation of the model. The dataset should include complete pharmacological and clinical drug-drug interaction data. Every encounter record should consist ofmedical care pairing details and unique identifiers for each medicine that interact similarly. On a scale from moderate to severe, encounter severity is harshness. The classification of interactions includes synergistic, antagonistic, and metabolic interference effects of drugs. Clinical outcomes include dizziness and a racing heart rate due to interaction-related side effects.In this study, compare the proposed ANFIS model's performance metrics—including Precision, Specificity, and Mean Squared Error (MSE)—to those of state-of-the-art models, including MCFF-MTDDI and DSIL-DDI, as well as the more conventional Reg-ML method. The experimental findings that were compared with each statistic are mentioned below: Precision(%) https://paperswithcode.com/dataset/ddi Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 92 https://internationalpubls.com One way to evaluate a model's performance is to look at its precision in Figure 4, which is defined as the percentage of predicted positive interactions. Necessary in DDI risk assessment for reducing the possibility of dangerous drug interactions being incorrectly labelled as safe, it assesses the model's ability to avoid false positives. Results for the Proposed ANFIS Model: 92.3%, MCFF-MTDDI: 88.1%, 85.7% for DSIL-DDI and 80.5% for Reg-ML. The results demonstrate that compared to other approaches, the ANFIS model performs better with a more excellent precision value. Fuzzy rules, which allow for more complex interaction evaluations and lessen the possibility of misclassifications, are responsible for this increase in the model. Fig. 4: Precision (%) Specificity (%) When assessing the specificity metric in Figure 5, one may look at how well the predicted negative interactions lined up with the actual ones.More specifically, it shows how the system can identify safe pharmaceutical combinations without misclassifying them. The Reg-ML model scored 83.8%, the MCFF-MTDDI model 90.2%, and the DSIL-DDI model 87.4%. The ANFIS model scored 94.6%, while the others scored and scored. The ANFIS model carefully selects safe pharmaceutical combinations for maximum specificity. This maximizes model efficacy. This high specificity shows that the model can reduce false positives in real-world DDI risk assessment, which is crucial. Fig. 5: Specificity Mean Square Error The MSE can calculate the average discordance between actual and anticipated values. Analyzing a model's predictive power this way is common. A low MSE suggests good performance. MCFF- MTDDI, Reg-ML, ANFIS Model, and DSIL-DDI estimate 0.056, 0.045, 0.034, and 0.021 on their 70 75 80 85 90 95 100 100 200 300 400 500 P re ci si o n (% ) No of images Reg-ML DSIL-DDI MCFF- MTDDI ANFIS model 0.7 0.75 0.8 0.85 0.9 0.95 1 100 200 300 400 500 S p e c if it y (% ) No of images Reg-ML DSIL-DDI MCFF- MTDDI ANFIS model Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 93 https://internationalpubls.com scales. The recommended ANFIS model predicts pharmaceutical interactions better than previous techniques. The MSE is substantially lower than previous approaches. Fuzzy inference and adaptive neural network learning make ANFIS powerful. We can map complex linkages more precisely because of this. Table 1 compares the Mean Squared Error (MSE) values of the proposed ANFIS model to those of other well-established DDI risk assessment approaches. Reg-ML, DSIL-DDI, and MCFF-MTDDI are cutting-edge models. Table 1: MSE Methods MSE ANFIS model 0.021 MCFF-MTDDI 0.034 DSIL-DDI 0.045 Reg-ML 0.056 Tests show that the DDI risk assessment using the recently established ANFIS model is more reliable and secure than the current one. The ANFIS model's low MSE and excellent specificity and accuracy allow it to discover safe and dangerous medication combinations. This reduces clinical medication interactions. 5. CONCLUSION This field is now assessing and predicting drunk driving risks using the ANFIS. Fuzzy logic and neural networks can help control medication interactions' complexity and unpredictability. The ANFIS model analyzes data using fuzzy rules and adaptive learning to assess DDI sensitivity. It's possible to seamlessly combine drugs, patient health, and pharmacological profiles. Experimental results show that the proposed model outperforms industry standards like MCFF-MTDDI and DSIL- DDI. Among these metrics are accuracy, specificity, and mean squared error. ANFIS outperforms rival models in error reduction and prediction accuracy due to its lower MSE score. This innovation is essential for pharmacological therapy management and patient safety. ANFIS-based systems may reduce clinically adverse drug interactions and increase medication safety. Because it accurately identifies high-risk interactions with few false positives. Further study may enlarge the dataset, integrate pharmacological components, and test the model in more clinical circumstances. The model's usability and practicability improve with this addition. REFERENCES [1] Hu, W., Zhang, W., Zhou, Y., Luo, Y., Sun, X., Xu, H., & Dai, H. (2023). MECDDI: clarified drug–drug interaction mechanism facilitating rational drug use and potential drug–drug interaction prediction. Journal of Chemical Information and Modeling, 63(5), 1626-1636. [2] Veera Boopathy, E., Peer Mohamed Appa, M.A.Y., Pragadeswaran, S., Karthick Raja, D., Gowtham, M., Kishore, R., Vimalraj, P., & Vissnuvardhan, K. (2024). A Data Driven Approach through IOMT based Patient Healthcare Monitoring System. Archives for Technical Sciences, 2(31), 9-15. [3] Zhong, Y., Zheng, H., Chen, X., Zhao, Y., Gao, T., Dong, H., & Weng, Z. (2023). DDI-GCN: drug-drug interaction prediction via explainable graph convolutional networks. Artificial Intelligence in Medicine, 144, 102640. [4] Neelima, S., Manoj, G., Subramani, K., Ahmed, A., & Chippy, M. (2024). Factors Influencing Data Utilization and Performance of Health Management Information Systems: A Case Study. Indian Journal of Information Sources and Services, 14(2), 146–152. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 2s (2025) 94 https://internationalpubls.com [5] Lv, Q., Zhou, J., Yang, Z., He, H., & Chen, C. Y. C. (2023). 3D graph neural network with few-shot learning for predicting drug–drug interactions in scaffold-based cold start scenario. Neural Networks, 165, 94-105. [6] Mohandas, R., Veena, S., Kirubasri, G., Thusnavis Bella Mary, I., & Udayakumar, R. (2024). Federated Learning with Homomorphic Encryption for Ensuring Privacy in Medical Data. Indian Journal of Information Sources and Services, 14(2), 17–23. [7] Zhu, J., Liu, Y., Zhang, Y., Chen, Z., She, K., & Tong, R. (2023). DAEM: Deep attributed embedding based multi- task learning for predicting adverse drug–drug interaction. Expert Systems with Applications, 215, 119312. [8] Camgözlü, Y., & Kutlu, Y. (2023). Leaf Image Classification Based on Pre-trained Convolutional Neural Network Models.Natural and Engineering Sciences, 8(3), 214-232. [9] Han, C. D., Wang, C. C., Huang, L., & Chen, X. (2023). MCFF-MTDDI: multi-channel feature fusion for multi- typed drug–drug interaction prediction. Briefings in Bioinformatics, 24(4). [10] Bobir, A.O., Askariy, M., Otabek, Y.Y., Nodir, R.K., Rakhima, A., Zukhra, Z.Y., Sherzod, A.A. (2024). Utilizing Deep Learning and the Internet of Things to Monitor the Health of Aquatic Ecosystems to Conserve Biodiversity. Natural and Engineering Sciences, 9(1), 72-83. [11] Khan, Mohammad Nazrul Islam, et al. "Geometric Properties of Submanifolds of a Riemannian Manifold in Tangent Bundles." Results in Nonlinear Analysis, vol. 7, no. 2, 2024, pp. 140–153. [12] Mansouri, S. (2023). Application of Neural Networks in the Medical Field. Journal of Wireless Mobile Networks, Ubiquitous Computing, and Dependable Applications, 14(1), 69-81. [13] Lu, Z., Suzuki, A., & Wang, D. (2023). Statistical methods for exploring spontaneous adverse event reporting databases for drug-host factor interactions. BMC medical research methodology, 23(1), 71. [14] Malathi, K., Shruthi, S.N., Madhumitha, N., Sreelakshmi, S., Sathya, U., & Sangeetha, P.M. (2024). Medical Data Integration and Interoperability through Remote Monitoring of Healthcare Devices. Journal of Wireless Mobile Networks, Ubiquitous Computing, and Dependable Applications (JoWUA), 15(2), 60-72. https://doi.org/10.58346/JOWUA.2024.I2.005 [15] Souza-Peres, J. V., Flores, K., Umloff, B., Heinan, M., Herscu, P., & Babos, M. B. (2023). Everyday Evaluation of Herb/Dietary Supplement–Drug Interaction: A Pilot Study. Medicines, 10(3), 20. [16] Cina, M., & Asha, P. (2024). FedProx: FedSplit Algorithm based Federated Learning for Statistical and System Heterogeneity in Medical Data Communication. Journal of Internet Services and Information Security, 14(3), 353- 370. [17] Sánchez-Valle, J., Correia, R. B., Camacho-Artacho, M., Lepore, R., Mattos, M. M., Rocha, L. M., & Valencia, A. (2024). Prevalence and differences in the co-administration of drugs known to interact: an analysis of three distinct and large populations. BMC medicine, 22(1), 166. [18] Kodric, Z., Vrhovec, S., & Jelovcan, L. (2021). Securing edge-enabled smart healthcare systems with blockchain: A systematic literature review. Journal of Internet Services and Information Security, 11(4), 19-32. [19] Gill, J., Moullet, M., Martinsson, A., Miljković, F., Williamson, B., Arends, R. H., & Pilla Reddy, V. (2023). Evaluating the performance of machine‐learning regression models for pharmacokinetic drug–drug interactions. CPT: Pharmacometrics & Systems Pharmacology, 12(1), 122-134. [20] Gupta, P., Majumdar, A., Chouzenoux, E., & Chierchia, G. (2023). DeConDFFuse: Predicting drug–drug interaction using joint deep convolutional transform learning and decision forest fusion framework. Expert Systems with Applications, 227, 120238. [21] Elamalki, Driss, Abdelilah Kaddar, and Nadia Beniich. "Stability of an Unemployment Model with a Non-Linear Job Creation." Results in Nonlinear Analysis, vol. 7, no. 2, 2024, pp. 127–139.