NARX-MLP: a hybrid model for accurate interpretable medical data classification

Machine learning plays a vital role in healthcare, yet medical datasets pose challenges such as nonlinear relationships, high-dimensional features, and the needs for model and result interpretability. We propose an adaptive NARX-MLP classifier, combining NARX with MLP and an adaptive feature selection procedure using L1 regularization. The performance (e.g., accuracy, precision, recall, and F1-score.) of the proposed methods is tested on two datasets: Hepatitis (static) and EEG Eye State (dynamic), to show the superiority of the new method. The selected features by this method can review nonlinear and temporal dependencies and therefore guarantee the capture of complex patterns while maintaining interpretability.