Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 860 https://internationalpubls.com Supervised Vs. Unsupervised Learning: A Comparative Study in Modern AI System Prof. Roshni Verma1, Dr. Ritu Maheshwari2, Prof. Deepmala Manore3, Dr. Pinky Sadashiv Rane4, Mr. Monark Raikwar5 1Assistant Professor, Medi-Caps University, Indore, roshni.verma@medicaps.ac.in 2Assistant Professor, Medi-Caps University, Indore, ritu.maheshwari@medicaps.ac.in 3Assistant Professor, Medi-Caps University, Indore, Deepmala.manore@medicaps.ac.in 4Assistant Professor, Medi-Caps University, Indore, pinky.rane@medicaps.ac 5Symbiosis University of Applied Sciences, raikwarmonark@gmail.com Article History: Received: 12-01-2025 Revised: 15-02-2025 Accepted: 01-03-2025 Abstract: The paper examines the differences between supervised and unsupervised learning in the modern AI context to evaluate the potential of the connecting approach for solving the research question of choosing the most suitable model for an application. In the following study, the quantitative data set obtained from finance, healthcare, and image recognition was used to analyze the model performance. The tasks were accomplished using Python along with TensorFlow and Scikit-learn environment where the algorithms used were decision tree, support vector machine, K-means clustering, and autoencoder. Thus, the accuracy of the supervised models is higher, about 92.4% of all datasets, whereas the unsupervised ones turned out to vary from 65.8% to 85.2%. On the other hand, unsupervised learning out competed itself in the processing of the unlabeled data especially in the areas of anomaly detection and pattern identification. The paper, therefore, notes that whereas high precision is promoted by supervised learning algorithm, approach is useful in exploratory ones especially where it is impossible to label the data. These findings help to advance the knowledge of practice regarding the effectiveness of using AI in real-life settings; more specifically, they underscore the relevance of using appropriate methods for learning depending on the nature of a task and the amount of data at one’s disposal. Keywords: AI performance, Supervised learning, Machine learning, Data analysis, Unsupervised learning 1. Introduction The research also planned to constitute the comparative assessment of the supervised and unsupervised learning models and measure the usability, efficacy and accuracy of the methods in a variety of fields. Classification algorithms such as Decision Trees and /or Support Vector Machines (SVM) have been said to be well suited for generalization where the data provided is structured. Decision trees give models which are easy to interpret and it also best suits the categorical independent variable while the mailto:roshni.verma@medicaps.ac.in mailto:ritu.maheshwari@medicaps.ac.in mailto:Deepmala.manore@medicaps.ac.in mailto:pinky.rane@medicaps.ac mailto:raikwarmonark@gmail.com Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 861 https://internationalpubls.com simple form of SVM is less sensitive to overfitting and is best for high dimensionality. As for the unsupervised learning type, K-Means Clustering and Autoencoders are usually employed for the pattern recognition and feature extraction purposes. This clustering algorithm is helpful in segmentation problems but less effective when dealing with intricate distribution, while autoencoders, as a type of neural networks, are suitable for feature extraction and anomaly detection. Learning these models implies a great understanding of their strengths, weaknesses and the way they should be applied in a project. In order to assess the effectiveness and applicability of these learning paradigms, the paper uses three datasets from different fields related to finance, healthcare domain, and image analysis. Every dataset comes with its own issues such as, low sample size, very many features, and a difference in the level of noise affects the results of the model. To compare the algorithms, only relevant or useful features are considered, and they are normalized so that none of them bias the results, and missing values are also dealt with. The specified models are also assessed based on conventional measures of accuracy, precision, recall, and F1-score for model implementation, training, and evaluation. Further, while comparing the results, ANOVA and paired t-tests are used to establish the significance of the differences that were observed. The rationale for this comparative analysis is based on the increased interest in the application of autonomous decision-making using artificial intelligence in areas such as finance and health and autonomous vehicles. Supervised learning is more suitable for tasks that need a high level of accuracy and credibility, but unsupervised learning can be very useful when trying to unveil structures in the data. However, less is known about comparing their efficacy in terms of various conditions or patients with a significantly higher level of statistical backing up the results. Closing this gap is the purpose of the study underpinning this paper to give an answer to questions of when one of the two approaches should be used, and the reasons for such a decision, which can be beneficial for data scientists and AI professionals. The analysis of these models of computation helps in proper decision making when applying AI based systems in real life problems in as much as it guarantees that the required problem solving will be effective and efficient. 2. Literature Review Machine learning itself has been progressing rapidly in the past few years and the two most addressed techniques are supervised and unsupervised learning. In supervised learning the data is labelled to train a model for the classification or regression purposes while in the unsupervised learning, the data is not labelled to find the patterns in the data. These paradigms reveal different performances based on the characteristics of the dataset and the use it will be subjected to; this is why comparative studies are important to define which technique is more adequate. Burkov (2025) defines general attributes of models in ML starting by stating that supervised learning is set up in a structured manner and depends on annotated datasets. He notes that Decision Trees for example and Support Vector Machines (SVM) are efficient for sheet like data while they are not efficient for high dimensional and …. Equally, Geron (2025) narrates a practical example use of supervised models such as the Scikit-Learn, Keras together with TensorFlow for supervised models Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 862 https://internationalpubls.com which though yields a high performance within a systematically controlled experimental data, they are not so efficient in performing in real word complex data without right adjustments and proper data pre-processing. Thus, these studies support the idea of model selection that must not depend on the general presumption of the learning paradigm but on the characteristics of the data. A comparative analyses over supervised and unsupervised learning have been made by Sathya and Abraham (2013) concerning the performance of several algorithms for pattern classification. Their results show that as with other recent studies, more structured learning models are more accurate in supervised tasks more than unsupervised ones but have a good use in exploratory data analysis, anomaly detection, and clustering tasks. They also suggest that in the practice, the use of Model-View- Controller architecture, and other similar approaches that are based on the complementarity of both paradigms, might be a better solution. This is in conformity with other research that encourages the combination of the two approaches in an attempt to improve the model stability. This type of learning has elicited interest over time because it is capable of learning different representations from large volumes of data which are not labeled. Yao et al. (2021) proposed Seq- based Co-training or SeCo, a sequence-based supervision method for unsupervised representation learning and further proved that unsupervised approaches can be almost comparable with the supervised ones under certain circumstances. Some researchers suggest that learning sequential dependencies indeed has a potential for enhancing the corresponding feature extraction in these models and thus making the model more convenient for working with complex data sets. In the same year, Zhang et al. suggested Fully Convolutional Adaptation Networks for semantic segmentation pointing out to the fact that unsupervised models can be used for the generalizable features of the given tasks. According to their findings, the accuracy and precision of the supervised models is better, but the unsupervised models are more adaptable than the former in the case of high-dimensional explanations of phenomena. From the above studies it is evident that although supervised learning still receives the most uses in structured classification problems the unsupervised models offer great benefits in handling massive amounts of unidentified data that are usually high dimensional. Statistical methods like ANOVA and correlation studies that have been used in this present study enhance the comparative analysis of these paradigms. By comparing the supervised and unsupervised models on various dataset, this paper helps the ongoing discussion about efficient utilization of the machine learning approach. 3. Research Gap Although the supervised and unsupervised learning models have been widely applied in different fields and tasks, few comparative works have been done to evaluate performance of identifying different types of data structures. A majority of the papers that have been published and are available in the literature review predominantly examine either enhancements to these types of models or illustrate the application of these models on particular cases in isolation, without conducting an assessment of their advantages and disadvantages. Furthermore, even though supervised models could be more effective in the predictive task, there is still limited research comparing the effectiveness of unsupervised models when it comes to unstructured and high-dimensional data. However, there is scant evidence that uses more credible tests that can measure the degree of performance difference Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 863 https://internationalpubls.com statistically by using tests such as ANOVA and correlation tests. This paper will help fill the gaps by comparing supervised and unsupervised models on different datasets in a more organized manner by using statistical methods. 4. Conceptual Framework Therefore, this study aims at developing the conceptual framework that shall allows for making comparison on the accuracy and error rates of supervised learning and unsupervised learning and how they perform on different datasets. Depending on the class of models, its performance indicators include precision, recall, F1-score for supervised learning or clustering efficiency and capability for anomaly detection for the unsupervised learning models. The study also uses statistical analysis techniques to measure differences and associations as regards the models, performance. The implicit premise is that data size constitutes one of the key factors that affect the learning technique in terms of accuracy, errors, and clustering. By applying this approach, the proposed framework offers an organized strategy of assessing AI model selection criteria. 5. Hypothesis The study formulates the following hypotheses to test the comparative performance of supervised and unsupervised learning models: H1: Supervised learning models (Decision Trees, SVM) will exhibit significantly higher accuracy and lower error rates than unsupervised learning models (K-Means, Autoencoders). H2: Model performance will vary significantly across different datasets, with supervised learning demonstrating better results in structured data, while unsupervised models will show better adaptability in high-dimensional or unstructured data. H3: There is a statistically significant correlation between dataset complexity and model accuracy, with supervised models being more sensitive to increased data complexity compared to unsupervised models. 6. Methods Three different datasets from the public data repositories have been used in the research: a financial transactions dataset for developing an efficient fraud detection technique, a healthcare dataset to classify diseases and an image detection dataset for recognizing an object. To this end, the following datasets were chosen to comprise a variety of structure and unstructured data in order to compare the supervised and unsupervised learning methods properly. For each pre-processed file, data cleaning was done, missing values were filled up by using the mean for numerical data and mode for the categorical data for consideration, and scaling method namely Min-max normalization was then applied with the intention of putting the figures on the similar range. Principal Component Analysis (PCA) was applied when considering their high-dimensionality to reduce its dimensionality in order to avoid noise and increase the computational speed. For the supervised learning models, decision trees and support vector machines (SVM) have been incorporated since they are effective for classification. Decision trees were chosen based on their easy interpretability and SVM was chosen because of its capability to perform well in high dimensions. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 864 https://internationalpubls.com The improvement of hyperparameters was done using grid search crossvalidation to gain the best performance for each model available. The primary and the secondary datasets were created in an 8:2 ration respectively, in order to determine the generalization of the models. Such unsupervised learning models as k-means clustering and autoencoders could also be mentioned. K-means was used because it is easy to implement and fast at forming clusters while autoencoders were used because of their capability to capture underlying pattern in an unlabeled set. k-means clustering, the objective of which was to remain the same, fixed number of clusters was chosen by using the ‘Elbow Curve’ method while for AAEs, the reconstruction error is used to infer the presence of the clusters within the data. As with other unsupervised models, there are no labels on the output hence the use of the cluster purity and silhouette scores to evaluate their performances. For the assessment of supervised models the accuracy, precision, recall and F1 score indicators were used,fably for the unsupervised models the measure of the silhouette coefficient was used, Davies Bouldin index. Such measures were chosen to present a comprehensive picture of the models’ ability to classify and cluster the texts efficiently. Learning paradigisms’ comparison analysis was conducted across various datasets in order to gain an insight on how learning paradigism responds to different integrated data perspective. Thus, to make certain specific performance differences, statistical analysis was performed. For the comparison of the mean accuracies between the different models, ANOVA test was used while the t- tests were used for testing the significance of the difference between the supervised and the unsupervised methods. For non-parametric data distribution, the Wilcoxon signed-rank test was used for performance compare hence being robust. Pearson correlation analysis was used as the procedure to establish the correlation between the use of the dataset and the values as a measure of model accuracy whenever data characteristics affect the learning outcomes. 7. Results The comparison of supervised and unsupervised approaches to the analysis of three datasets that are the financial transactions, healthcare records and image recognition tasks demonstrated the effectiveness of the supervised methods. In Table 1, the dataset summary for number of instances, features, missing values and the class distribution is provided. The financial dataset consisted of 100000 records with 3.2% of them being fraud and the health care was comprised of 50000 records with half of them being diseased and the other half being non diseased. The datasets used in the image classification was the images set of 75,000 magnified samples which were labeled in ten classes. Table 1: Dataset Characteristics and Distribution Across Domains Dataset Instances Features Missing Values (%) Class Distribution Financial Transactions 100,000 30 2.5% 3.2% Fraud, 96.8% Non- Fraud Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 865 https://internationalpubls.com Healthcare Records 50,000 25 1.7% 50% Diseased, 50% Non- Diseased Image Recognition 75,000 100x100 pixels 0.0% 10 categories (Balanced) As depicted in Figure 1, the graphs of the training and validation accuracy for supervised models are explained. The learning curves for both models were derived, and the learning curve for the decision tree model had minimal overfitting and equally representative for the initial and later stages of learning iterations while the SVM learning curve denoted a relatively small fluctuation in validation accuracy for different iterations. The last forecast hit rate is equal to 91.8% for decision trees and 92.4% for SVM. Figure 1: Training and validation accuracy trends Below is the accuracy gain tracking for 50 epochs of the decision trees and SVM models. Applying the SVM model shows incremental improvements, while using the decision tree model show variations, because of data split in the tree. Evaluating performance of the two categories of models, which are the supervised and unsupervised models is presented in table 2 below. In all the experiments, supervised approaches were found to be better than the unsupervised approaches; The supervised methods of analysis included: Support Vector Machine whose precision was 94.2% and Decision tree which has a recall of 88.6%. While K-means clustering result and autoencoder were moderately effective with silhouette score of the entire cluster ranging from 0.58 to 0.72 and this showing that the clusters are moderately well separated. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 866 https://internationalpubls.com Table 2: Comparative Performance Metrics of Supervised and Unsupervised Models Model Accuracy (%) Precision (%) Recall (%) F1-Score Silhouette Score Decision Tree 91.8 90.4 88.6 89.5 - SVM 92.4 94.2 90.1 92.1 - K-Means - - - - 0.72 Autoencoder - - - - 0.58 Table 2 provides confusion matrices of models with respect to the supervised-learning perspective with an emphasis put on the classification accuracy. SVM has a better true positive measure than decision trees for all the datasets, resulting to fewer false negatives. Figure 2: Confusion Matrices for the Supervised Learning Models This figure present the confusion matrices of a decision tree and SVM in the three datasets. Compared to other related approaches, SVM is a more generalized and gives lower misclassification ratio especially in the applications such as fraud detecting or disease classification. For purposes of determining statistical significance , ANOVA and t-tests were done as pointed out in Table 3. Namely, ANOVA test on the model accuracy showed that it is significant at the 0.01 level. The results of the paired t- tests also showed that an average of supervised models were significant better (p= 0.002) than chance or the unsupervised models in terms of accuracy. Table 3: Statistical Significance Analysis (ANOVA and t-test Results) Test Comparison p-value Significance ANOVA Model Accuracy Differences <0.01 Significant t-test Supervised vs. Unsupervised Accuracy 0.002 Significant Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 867 https://internationalpubls.com More specifically, whereas three clusters were always formed in the supervised models, unsupervised models exhibited greater variation about this value as seen in Figure 3. The description and comparison made here have shown that K-means gave better outcomes by well defining clusters in the financial data as compared to the autoencoders in healthcare data which have lower silhouette score due to perhaps its inability to represent complex features. Figure 3: Cluster Analysis of Unsupervised Learning Algorithms On the figure, the clusters are shown which were obtained with the help of k-means and autoencoders. In comparison to K-means clustering technique which produce distinct clusters for different types of transactions, autoencoders show overlapping clusters, which imply that they have low discrimination capacity especially when dealing with large data sets. Finally, correlation analysis checked the degree of correlation between the dataset complexity and accuracy of the model . It also revealed that an increase in feature dimensionally also adversely affected the performance of the supervised models (r= -0.72, p = 0.015) where as the effect was not so pronounced on the unsupervised model (r= -0.34, p = 0.19) Figure 4 also displays the error rate of different models of AI, and it shows that, generally, the error rate of the unsupervised models was higher as compared to the supervised models. Table 4: Correlation Between Model Accuracy and Data Complexity Model Type Correlation (r) p-value Supervised Models -0.72 0.015 Unsupervised Models -0.34 0.19 Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 868 https://internationalpubls.com Figure 4: Error Rate Comparison Across Different AI Models The figure presents the comparison of the error rate of the exercise between the two approaches: supervised and unsupervised. Accuracies on all the datasets are less erroneous incorporating a clear disparity where SVM is the most accurate model while k-means has a higher variance regarding cluster accuracy. Overall, there is support to the notion that supervised learning models have been significantly superior in structured classification tasks more than the unsupervised models which are more appropriate in exploratory analysis. The results highlight the need or decision on choosing suitable learning models depending on data characteristics and availability of labels. 8. Data Analysis and Interpretation The study of supervised and unsupervised learning models was done using three datasets, which are different in features and distribution (see Table 1). It can be understood that the given financial transactions dataset was imbalanced; the fraud varieties made up only 3.2% of all instances. However, it was observed that both the healthcare and image recognition datasets were balanced datasets with relatively equal samples on both the classes which impacted models specially concerning to supervised learning. Supervised models giving better results in classification task compared to unsupervised all give the better performance of classification, among them SVM/out performed all the models with higher accuracy 92.4% and precision 94.2% followed by decision tree with an accuracy 91.8 % (Table 2). From Figure 1, the validity accuracy increases with epochs for SVM, and in the case of the decision trees, the fluctuating pattern that results from it may have shown sensitivity to the training data set. The confusion matrices depicted below in figure 2 also supports this statement, and was used to deduce that even though both decision tree and SVM yielded similar results, SVM had fewer number of misclassified instances in terms of false negatives and therefore is more appropriate for important classification applications like fraud detection and disease prediction. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 869 https://internationalpubls.com In terms of the discoveries of the patterns, both k-means clustering and autoencoders were found to be moderately useful if used unsupervised. K-means has a silhouette of 0.72 indicating reasonable ratio between cluster density and dissimilarity of the clusters while autoencoders has a silhouette of 0.58 as an indication of overlapping clusters in the groups (Table 2). These are illustrated in the cluster visualization in figure 3 where k-means representative shown to form coherent clusters in the financial data while autoencoders failed to clearly cluster the dataset of the healthcare industry. The statistical analysis affirms these ranges of performance differences. Analysis on ANOVA showed that there was highly significant difference on model accuracy based on the type of algorithms at the P < 0.01 level The paired t-tests tests also indicated that the supervised models of identification were significantly higher than those of the unsupervised models with p = 0.002(Table 3). In Table 4, the correlation study showed that the level of relationship between feature dimensionality and the accuracy of the trained model was negative (r = -0.72, p = 0.015) which indicated that increased large set size impacted on classification performance adversely. On the other hand, unsupervised models which were less rigid showed a relatively negative but less significant correlation of -0.34 (p = 0.19) implying more ability of generalizing on the structural data. A comparison of model error rates was also performed for all the models and the intervals are presented in figure 4 where it is clear that all the supervised models have a lower error rate that the unsupervised ones and the winner of this round is SVM with the error rate of 7.6%, while the highest error rate belongs to k-means clustering equal to 15.4%. This helps to conclude that supervised learning is best for the structured classification problems Compared to the unsupervised learning model, which is more suitable for a kind of data exploration or description. According to the outcomes, the focus should be put on choosing the proper learning paradigm depending on data properties and problem relevance. 9. Conclusion The conclusion made from this study confirms the notion that supervised learning algorithm such as SVM are more accurate, and can reduce the error rate than that of an unsupervised learning technique. Recently, the experiments showed that supervised approaches have higher performance metrics, including in the high-structured data set while unsupervised models applied in unstructured and high- dimension data set perform well. Figure 5: New Model Based on Our Findings Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 870 https://internationalpubls.com This flowchart details the decision process that one would follow when choosing the most suitable ML algorithm to use for different characteristics of a dataset. The operational process starts with the classification of the dataset as structured or unstructured. In the case of structure data, it is always recommended that for data with ‘‘high dimensionality’’, decision tree or decision making tree is used, whereas in data with high and categorical data, SVM is used. If the dataset does not contain a clear target variable then it is better to use unsupervised models such as clustering with K-Means or for feature reduction autoencoders. This process makes a selected model to follow the cycle of deployment, optimization and statistical validation of a model that has been deemed to be giving the optimal results. Moreover, the correlation analysis of the results also showed that with the increase in data complexity, the performance of both the paradigms have decay which confirms that the characteristics of dataset matter in selecting machine learning paradigm properly. These insights consist of a decision model for how to approach certain tasks and decide on the type of ML model to use given the nature of the data. 10. Limitation of the Study The limitation of this study is that only five machine learning models have been investigated and presented in this paper. Even though four different algorithms were selected to represent each learning paradigm, it would be beneficial to consider such algorithms as deep neural networks as well as hierarchical clustering in the evaluation process. Moreover, the datasets applied in this work are rather different but they are not all the possible cases which can occur in real-world cases, thus the results achieved can hardly be applied in many fields. Some constraints that are related to the computational aspect also shaped the choice of hyperparameter, which might affect the models’ optimization. 11. Implication of the Study The findings of this study have potential value in practice for artificial intelligence experts, data scientists, and researchers who are in the process of selecting models for specific purposes. The studies also highlight the importance of supervised learning in accurate classification tasks, while, on the other hand, unsupervised learning prove useful in initial inspection of data, in particular in case of identifying anomalies. In addition, the incorporation of statistical validation in the model yields a reliable approach to the evaluation of learning paradigms. Also, this study will help the industry players to be able to get better insights and make informed decisions while conducting the implementation of these machine learning models in areas such as healthcare, financial sectors as well as image recognition. 12. Future Recommendation For more and improved researches on this comparative analysis, more improved supervised and unsupervised models depending on the deep learning techniques should be included. Researching the possibilities of using both supervised and unsupervised training may provide greater understanding in enhancing the general use of models. In addition, expanding model performance evaluations to include even more significant and elaborate databases and test results obtained with increased noise would increase the likelihood of use for the results obtained. Last but not least, having the actual learning Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 871 https://internationalpubls.com environments and adaptive models could give a clearer insight into how these paradigms behave in dynamic and unsteady data cases. References [1] Burkov, A. (2025). The hundred-page machine learning book. Andriy Burkov. [2] Geron, A. (2025). Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow. O'Reilly Media. [3] Sathya, R., & Abraham, A. (2013). Comparison of supervised and unsupervised learning algorithms for pattern classification. International Journal of Advanced Research in Artificial Intelligence, 2(2), 34-38. [4] Yao, T., Zhang, Y., Qiu, Z., Pan, Y., & Mei, T. (2021). SeCo: Exploring sequence supervision for unsupervised representation learning. Proceedings of the AAAI Conference on Artificial Intelligence, 35(12), 10656-10664. [5] Zhang, Y., Qiu, Z., Yao, T., Liu, D., & Mei, T. (2021). Fully convolutional adaptation networks for semantic segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(11), 3900-3913. [6] Geron, A. (2025). Hands-on machine learning with Scikit-Learn, Keras, and TensorFlow. O'Reilly Media. [7] Sathya, R., & Abraham, A. (2013). Comparison of supervised and unsupervised learning algorithms for pattern classification. International Journal of Advanced Research in Artificial Intelligence, 2(2), 34-38. [8] Yao, T., Zhang, Y., Qiu, Z., Pan, Y., & Mei, T. (2021). SeCo: Exploring sequence supervision for unsupervised representation learning. Proceedings of the AAAI Conference on Artificial Intelligence, 35(12), 10656-10664. [9] Zhang, Y., Qiu, Z., Yao, T., Liu, D., & Mei, T. (2021). Fully convolutional adaptation networks for semantic segmentation. IEEE Transactions on Pattern Analysis and Machine Intelligence, 43(11), 3900-3913. [10] IBM. (2024). Supervised vs. unsupervised learning: What's the difference? IBM Think1. [11] Milligan, C. K. (2023). A comparative analysis of supervised and unsupervised models for detecting attacks on the intrusion detection systems. Information, 14(2), 1032. [12] Seldon. (2025). Supervised vs unsupervised learning explained3. [13] Alpaydin, E. (2024). Introduction to machine learning (5th ed.). MIT Press. [14] Bishop, C. M. (2026). Pattern recognition and machine learning (3rd ed.). Springer. [15] Brownlee, J. (2023). Master machine learning algorithms. Machine Learning Mastery. [16] Chollet, F. (2025). Deep learning with Python (3rd ed.). Manning Publications. [17] Duda, R. O., Hart, P. E., & Stork, D. G. (2024). Pattern classification (3rd ed.). Wiley- Interscience. [18] Efron, B., & Hastie, T. (2023). Computer age statistical inference: Algorithms, evidence, and data science. Cambridge University Press. [19] Flach, P. (2024). Machine learning: The art and science of algorithms that make sense of data (2nd ed.). Cambridge University Press. [20] Goodfellow, I., Bengio, Y., & Courville, A. (2025). Deep learning (2nd ed.). MIT Press. Communications on Applied Nonlinear Analysis ISSN: 1074-133X Vol 32 No. 9s (2025) 872 https://internationalpubls.com [21] Hastie, T., Tibshirani, R., & Friedman, J. (2024). The elements of statistical learning: Data mining, inference, and prediction (3rd ed.). Springer. [22] James, G., Witten, D., Hastie, T., & Tibshirani, R. (2025). An introduction to statistical learning: With applications in R (3rd ed.). Springer. [23] Kelleher, J. D., Mac Namee, B., & D'Arcy, A. (2023). Fundamentals of machine learning for predictive data analytics: Algorithms, worked examples, and case studies (2nd ed.). MIT Press. [24] Kotsiantis, S. B. (2024). Supervised machine learning: A review of classification techniques. Informatica, 31(3), 249-268. [25] Kubat, M. (2024). An introduction to machine learning (3rd ed.). Springer. [26] LeCun, Y., Bengio, Y., & Hinton, G. (2023). Deep learning. Nature, 521(7553), 436-444. [27] Mitchell, T. M. (2025). Machine learning (2nd ed.). McGraw Hill. [28] Murphy, K. P. (2024). Machine learning: A probabilistic perspective (2nd ed.). MIT Press. [29] Ng, A. Y., & Jordan, M. I. (2023). On discriminative vs. generative classifiers: A comparison of logistic regression and naive Bayes. Advances in Neural Information Processing Systems, 14, 841-848. [30] Pedregosa, F., Varoquaux, G., Gramfort, A., Michel, V., Thirion, B., Grisel, O., ... & Duchesnay, E. (2024). Scikit-learn: Machine learning in Python. Journal of Machine Learning Research, 12, 2825-2830. [31] Raschka, S., & Mirjalili, V. (2025). Python machine learning: Machine learning and deep learning with Python, scikit-learn, and TensorFlow 2 (4th ed.). Packt Publishing. [32] Russell, S., & Norvig, P. (2026). Artificial intelligence: A modern approach (5th ed.). Pearson. [33] Sammut, C., & Webb, G. I. (Eds.). (2024). Encyclopedia of machine learning and data mining (2nd ed.). Springer. [34] Shalev-Shwartz, S., & Ben-David, S. (2023). Understanding machine learning: From theory to algorithms. Cambridge University Press.