Cardiovascular disease risk prediction using machine learning techniques

Abstract

Cardiovascular disease (CVD) is one of the leading cause of deaths across the globe. Several indicators can help identify the risk of developing CVD that include both physiological and behavioral aspects. However, there is no such mechanism that can determine which are the most effective indicators for such a prediction. Nonetheless, with the advancement in Machine Learning and Data Mining techniques, risk prediction systems for CVD can be developed. In this thesis, we present an in-depth analysis of clinical data using various machine learning techniques to develop a risk prediction system that can provide auxiliary medical assistance to experts. By employing a publicly available database, we identify the most effective features using feature selection. Main challenge in the analysis of medical data is the unstructured or semi-structured nature of the available information that can adversely impact the prediction. To enhance prediction data cleaning and pre-processing is employed. For the classification purposes, a hierarchical approach is adopted where the first classifier discriminates between the healthy and patient data and the second rates severity of the disease. A detailed empirical analysis of the proposed scheme validates the performance of our technique