Optimization of Deep Learning Approach Using Chaotic Algorithm for Hepatitis Detection

Hepatitis still remains one of the major health issues in the world and is responsible for a significant amount of morbidity and mortality due to diseases related to the liver. Traditional approaches for hepatitis detection are costly, laboratorial, time-consuming, and error-prone. While AI and DL have enhanced medical diagnosis significantly, current models are prone to problems such as overfitting, long convergence period, and lack of generalization, which are primarily associated with inefficiency of hyperparameter optimization. In this research, the authors propose the use of optimized deep learning model based on Chaotic Optimization Algorithm (COA) for hepatitis detection. This approach combines Multi-Layer Perceptron (MLP) and Deep Neural Network (DNN) models with the application of COA, which allows optimizing such parameters as learning rate, hidden neurons, momentum, and weights initialization. Two publicly available datasets (the Kaggle Hepatitis Dataset and the UCI Hepatitis Dataset) have been used in this research. Data preprocessing steps including handling missing values, encoding, normalization, and splitting data into train and test datasets were applied. The COA-DNN gave results of 85.11% for accuracy, 77.83% for precision, 83.24% for recall, 79.85% for F1 score, and 0.9054 for AUC on the Kaggle dataset compared to the conventional DNN which had an accuracy of 57.45%. For the UCI dataset, the highest accuracy rate was achieved by COA-DNN at 96.76%, precision of 95.86%, recall of 88.82%, F1 score of 91.94%, and AUC of 0.9793.

Keywords: Artificial Intelligence; Chaotic Optimization Algorithm; Deep Learning; Hepatitis Detection; Hyperparameter Optimization; Multi-Layer Perceptron; Neural.