Chi-Square and Cuckoo Search Based Feature Selection for Heart Disease Prediction

In many countries across the globe, heart disease is the main root of death. It’s only that it has a lot of features that have been validated by numerous sources and incorporate distinct diagnostic techniques. Thus, physicians find it difficult to correctly identify the features that contribute to heart disease for each patient. Besides, most of these features are redundant and irrelevant that create noise towards the description of the target class. Nowadays, machine learning approaches are being used to aid physicians to correctly identify heart disease from a patient. In this study, the chi-square statistic is applied to rank the attributes in the heart disease data set while the cuckoo search optimization algorithm selects the most useful features that contribute significantly to heart disease diagnosis. The selected features are trained using the k-nearest neighbour (KNN) along with support vector machine (SVM). The investigational outcomes displayed that the features are minimized to the lowest level, mainly on the high dimensional data sets. Therefore, contributed positively to the heart disease diagnosis system. SVM outpace KNN in terms of accuracy after the proposed feature selection. On the other hand, KNN performed slightly better than SVM without feature selection on the majority of data sets. The proposed method performed better than other existing approaches especially correlation-based FS with PSO together with fuzzy analytic hierarchy process-based feed-forward neural network, modified differential evolution, on both percentage accuracy and number of chosen features (feature size).