Improving the Classification Performance for the Effective Diagnosis of Diseases Using Optimized Weighted KNN on Clinical Dataset

Background: This study introduces an effective technique for diagnosing diseases more accurately using the improved machine learning algorithm on clinical datasets. Early detection and prediction of diseases can significantly improve survival rates. Objective: The aim of this paper is to propose a novel approach to enhance the conventional KNN performance for predicting diseases more precisely. Method: Optimized feature and distance weights and Bayesian inspired refinements are employed in the proposed approach which is evaluated by conducting experiments on five widely-used UCI and Kaggle datasets and compared with 9 contemporary variants of KNN as well as the traditional KNN. The accuracy, precision, and recall measures are used for performance comparison. Findings: This research highlights the potential of machine learning, particularly the proposed OWKNN, in enhancing medical diagnostics. Improving the performance of the machine learning algorithms for medical diagnosis enables early detection and prompt intervention, thereby potentially improving patient outcomes. Novelty: The improved version of the KNN algorithm has been invented by integrating the feature, distance weights optimized through Genetic Algorithm and refined with the Bayesian framework. The findings demonstrate the usefulness of the OWKNN in accurately predicting diseases and offering a more reliable avenue for enhancing precise detection strategies. Result and Discussion: Experimental results show that the proposed method consistently outperformed the traditional KNN and considered variants of KNN in terms of average accuracy, precision, and recall scores of 91.79%, 93.16%, and 88.86% respectively across various datasets. Thus, this work comes up with an improved version of the machine learning algorithm (KNN) for effective medical diagnostics. Keywords: KNN Classification, Optimized-Weighted KNN, Mahalanobis Distance, Mutual Information, Bayesian framework, Genetic Algorithm