This paper presents more accurate and reliable computational methods for aiding the treatment of people with coronary artery disease. New techniques are introduced for improved evaluation and distinguish cardiac disease affected patients from the healthy controls. Experiments are conducted with high level of error tolerance rate and confidence level at 95% and 99% and established the results with corrected T-tests based on comparison of various performance measures. Normal kernel density estimator is used for visual distinction of cardiac controls. A new ensemble learning method comprising of Bayesian network as classifier and Principal components method as the projection filter with ranker search is used for the relevant feature selection. Analysis of each model is performed and discusses major findings and concludes with promising results compared to the related works. Multiple Correspondence analysis is used for exploring heart disease variable's relationships. Robust machine learning algorithms used are Rotation forests, MultiBoosting, Sparse multinomial logistic regression for better performance with fine tuning of their involved parameters. The work aims at improving the software reliability and quality of diagnosis of cardiac disease with robust inference system. To the best of our knowledge, from the literature survey, experimental results presented in this work show best results with supportive statistical inference.