Prediction of heart disease dataset using hybrid optimization and machine learning techniques

Abstract

The traditional methods of cancer diagnosis and cancer-type recognition have quite a large number of limitations in terms of speed and accuracy. However, recent studies on cancer diagnosis are focused on molecular level identification so as to improve the capability of diagnosis process. Several recent research studies have used data mining techniques, machine learning algorithms and statistical methods to study the issue of cancer classification to achieve an effective analysis on gene expression profiles. The process has been supported by data mining and machine learning techniques such that a smart combination of various algorithms and techniques can generate comparatively efficient and reliable output. In this thesis, four distinct classification methods are used – Support Vector Machine (SVM) (SVM with Gaussian Kernel and SVM with Radial Basis Kernel), K – Nearest Neighbor, Naïve Bayes and Random Forest to study the accuracy of prediction using heart disease dataset. The models were trained using different training-testing ratios to get the optimal results. Three optimization techniques were used to further improve the prediction accuracy of heart disease. Various performance parameters including accuracy and area under the ROC curve (AUC) were computed to test the optimization algorithms, Grey Wolf Optimization (GWO), Oppositional Grey Wolf Optimizer (OGWO) and Hybrid Particle Swarm Optimization with Grey Wolf Optimizer (HPSOGWO) with the four classifiers to obtain the best optimized classifier for the heart disease datasets. The datasets used in this thesis are of heart disease taken from three different locations: Cleveland, Hungarian and Switzerland. The results show that best accuracy was achieved by using the hybrid PSO with GWO (HPSOGWO) in Switzerland (with Random Forest) and Hungarian (with KNN) datasets whereas GWO gave the highest accuracy for the Cleveland (with Random Forest) dataset with the HPSOGWO very close in accuracy to GWO.