PREDICTING CORONARY HEART DISEASE USING AN IMPROVED LIGHT GBM MODEL PERFORMANCE ANALYSIS AND COMPARISON

Abstract

Unfortunately coronary heart disease (CHD) is a serious heart condition that is a serious health threat and has no "cure". Coronary artery disease can be detected early and accurately to provide optimal care to patients. With timely identification, interventions and better outcomes for the patient are possible. To predict CHD with a LightGBM classifier, the suggested “HY_OptGBM” model is intended to use an optimized LightGBM. LightGBM is an efficient and accurate predictive modeling framework, which is highly effective on gradient boosting. The hyperparameters of the LightGBM classifier is tuned with additional improvement of the loss function to fine tune the LightGBM classifier. This optimization technique contributes to the model training process, which in turn improves the accuracy and efficiency of the model. The model is evaluated on the data of the Framingham Heart Institute for coronary heart disease. This data helps the model to predict CHD well and so, can diagnose CHD early and possibly minimize the cost of treatment by treating the illness in its early stages. To increase the identification accuracy of CHD, and to suggest Voting Classifier (RF + AdaBoost) with an excellent accuracy of 99%. The ensemble model of Random forest and Ada Boost shows robustness in distinguishing the patterns related to CHD. A user-friendly Flask structure connected to SQLite is provided, making the use of it easy and simplifying the user test signin and signup procedure. This simplified interface enhances the accessibility, thereby making the ML approaches more practical and user friendly for many stakeholders involved in CHD diagnosis.