Optimal Prediction of Cardiovascular Disease Using Stochastic Layer-wise Autoencoder

Abstract

Today, the world's leading cause of death is cardiovascular disease. Based on medical records, the machine learning approach has been used in some studies to predict cardiovascular diseases. However, there is still a need for an optimal approach because medical records have a high correlation with data. Sometimes, data with high dimensions leads to inefficiency. An effective dimensionality reduction technique is necessary to reduce noise and high-dimensional data. This paper proposed a novel stochastic layer-wise sparse autoencoder model with a deep feed-forward neural network (SLSAE-DFFN). Unlike traditional sparse autoencoders, which use a fixed sparsity across the entire network, our method introduces probabilistic layer-wise sparsity. This approach reduces overfitting and improves generalization with diverse latent feature representations. After training, lower-dimensional features from the bottleneck layer of the autoencoder are used to train a deep feedforward neural network for classification. This proposed model shows great promise in predicting cardiovascular disease, achieving an impressive 97% accuracy. The SLSAE-DFFN model is carefully evaluated using various performance metrics, including accuracy, precision, recall, and the F1-score