Multimodal deep learning for chronic kidney disease prediction: leveraging feature selection algorithms and ensemble models

Abstract

ABSTRACT This research presents an advanced approach to enhance disease diagnosis using imbalanced medical datasets. Feature selection techniques, LASSO and Relief, are applied to identify relevant features from the UCI dataset and missing values are handled appropriately. To address the class imbalance, SMOTEENN is used, creating a new combined dataset with selected features. Three deep learning models, FNNs, LSTMs, and GBMs, are employed and trained on the combined dataset, achieving remarkable accuracy (1.0). Evaluating the models on LASSO and Relief datasets independently, FNN/MLP obtains perfect accuracy, GBM performs well (0.9888 on LASSO and 1.0 on Relief), and LSTM shows good results (0.9663 on LASSO and 1.0 on Relief). This study demonstrates the effectiveness of combining LASSO and Relief for feature selection and highlights the impact of SMOTEENN on model performance. The achieved accuracy with all models on the combined dataset showcases deep learning's potential for accurate disease diagnosis even with imbalanced data, offering promising insights for robust medical diagnosis systems.