Comparative Analysis Of Machine Learning Techniques For Applications In Effective Decision Making

Abstract

Data mining has now become an unavoidable field in modern information technology era, contributing considerably to decision-making across diverse fields such as health care, finance, marketing, and many others. The key essence of data mining is that it discovers unknown patterns, meaningful insights, and relationships within very large datasets so that predictive analytics and informed decisions can be developed. This thesis explores new methods to improve the accuracy and efficiency of data mining processes, particularly in the application of classification and clustering techniques. newlineThe research involved in this thesis focuses on developing and evaluating some advanced methodologies related to the application of fuzzy logic, hybrid models, and some novel loss functions to overcome obstacles in classification and clustering. To this end, a new loss function called I2CS Loss (Intra-Concentration and Inter-Separability) has been proposed; it is also integrated into the Neighbourhood Component Analysis (NCA) to make feature selection more effective and improve clustering. The use of fuzzy logic techniques, therefore, also ensures adaptability and robustness in dealing with real-world uncertainties inherent in datasets. newlineA significant portion of the work will be devoted to the experimental assessment of the suggested models on generally known datasets like Wisconsin Diagnostic Breast Cancer. The outcome showed that the presented methodologies are better in terms of accuracy, precision, recall, and overall classification performance compared with traditional approaches. The introduction of new metrics and optimization techniques makes the clustering and feature selection robust. newlineThis thesis is designed to comprehensively explore the theoretical underpinnings, experimental design, and practical implications of the proposed methods. It begins with a foundational understanding of data mining and its challenges, followed by a detailed examination of the developed algorithms, their mathematical formulations, and their inte