Hyperparameter Tuning Based Performance Analysis of Machine Learning Approaches for Prediction of Cardiac Complications

Abstract

Nowadays, cardiac disorders are the biggest causes for morbidity and mortality in the world causing huge number of deaths over the last few decades and have emerged as the most life-threatening disease. So, this calls for immediate need of reliable, accurate and feasible system to diagnose and predict such diseases for timely diagnosis. This paper emphasizes on different approaches using classical machine learning algorithms with their hybrid approaches for the predictions of heart attacks, so as to compare their performance. Here, classical and hybrid machine learning algorithms and techniques have been applied to the standard medical data sets to come out with lab-based prototype in an attempt to automate the analysis of large and complex patient data attributes. Comparative performance analysis of different scalar has been carried out to use standard scalar for feature scaling method. It initially deals with approach pertaining to classical models based on supervised learning algorithms such as Support Vector Machines, K-Nearest Neighbor, Random Forest and Deep learning models. Hybrid approach using Fuzzy C means based Neural Network gives classification accuracy of 98.76% which shows that this proposed optimized technique outperforms much better over the simple classifiers in terms of prediction accuracy. Using huge medical data in health care systems with validated and appropriate data splitting for training as well as testing, attempts have been made to compare performance of different algorithms for different ratios of data used for training and testing to fine tune accuracy requirement to understand the impact of these ratios. This work envisages GridsearchCV to find the best suitable values for the hyper parameters and its impact on improvising accuracy of algorithmic models leading to more accurate prediction of heart complications.