Efficient Heart Disease Prediction Using Hybrid Deep Learning Classification Models

Abstract

INTRODUCTION: Heart disease (HD) has been identified as one of the deadly diseases, which affects the human beings of all ages worldwide. In such a scenario, Data Mining (DM) techniques have been found to be efficient in the analysis and the prediction of the phases of HD complications while handling larger patient datasets'. This dataset would consist of irrelevant and redundant features as well. These features would further impact the accuracy and the speed of data processing during the classification process.

OBJECTIVES: Hence, the feature selection techniques are required for removing the redundant features from the dataset. Therefore, in this study, feature selection techniques like genetic algorithm, particle swarm optimization and African buffalo algorithm have been implemented.

METHODS: To further enhance this process, a newly developed GSA (Genetic Sine Algorithm) is proposed as it is capable of selecting optimal features and avoid getting trapped in local optima. The selected features are subjected to the classification technique by RNN (Recurrent Neural Network) integrated with LSTM (Long Short Term Memory) algorithm. To filter out all the invalid informations and emphasize only on critical information, DPA-RNN+LSTM (Deep Progressive Attention-RNN+LSTM) has been developed so as to improve the classification rate.

RESULTS: The proposed results have been supported by the performance and comparative analysis performed on two benchmark datasets namely heart disease diagnosis UCI dataset and heart failure clinical dataset. Further, statistical analysis in terms of Mann-Whitney U-test, Pearson Correlation co-efficient, Friedman rank and Iman-Davenport significant values has been evaluated.

CONCLUSION: The obtained results show that the proposed system is comparatively more efficient for heart disease diagnosis than other conventional techniques.