A Study on the Extended Hybrid Feature Selection Approach with Dimensionality Reduction for Improved Classification

Tan, Zi Xuan (2022) A Study on the Extended Hybrid Feature Selection Approach with Dimensionality Reduction for Improved Classification. Masters thesis, Tunku Abdul Rahman University of Management and Technology.

Text 15 TAN ZI XUAN (MCS).pdf Restricted to Registered users only Download (6MB)

Abstract

Heart Disease has become the major cause of death around the world. The World Health Organization has recorded around 17.9 million deaths caused by cardiovascular heart disease (CVD). This is popular among the low and middle-income population, where resources and benefits of healthcare programs are lacking and people are not able to pay for the expensive procedures. Hence, having an efficient and effective solution in detecting heart disease occurrence is crucial. In this research, a novel approach of hybrid feature selection with injection of feature dimensionality was studied. Where the proposed method combines the strength of different feature selection techniques and enhanced by dimensionality reduction techniques. This research has used various sources of heart disease dataset from Cleveland, Framingham, Z-Alizadash and Cardiovascular Dataset. The objective is to improve classification performance using the proposed hybrid feature selection engine and study the potential of generalization of the proposed method. The performance of the feature subsets will be evaluated using various machine learning models. The model performance will be compared and studied using accuracy. With the proposed engine, classification performance was improved up to 20% when applied with feature subset produce by the proposed engine. Random Forest () and K-Nearest Neighbour both have the most improvement observed. Overall, highest classification achieved were accuracy (0.89), precision (0.94) and f1-score (0.92).