Applying feature selection and weight optimization techniques to enhance artificial neural network for heart disease diagnosis

Abstract

Heart diseases can result in certain death, if not diagnosed timely. Heart diseases have caused death irrespective of age, gender or any other demographics. Its diagnosis is, hence, very significant, and angiography is used most commonly for the stated purpose, but it is very expensive and causes side effects. Machine learning techniques are used for diagnosing heart diseases. This study utilizes Z-Alizadeh Sani heart disease dataset for experimenting four feature selection and four optimization techniques to improve the performance of a standard neural network. The experimentation has helped us in finalizing the proposed technique, and the final technique has also been applied on two other datasets in order to ensure that overfitting is avoided. The results of the final application are promising, and in the future, we intend to form a subset of those features which are selected by all the techniques. We also intend to make an ensemble of the best weight optimization techniques in order to further improve the standard neural network.