APPLICATION OF CONVOLUTIONAL NEURAL NETWORKS (CNN) FOR HEPATITIS C VIRUS (HCV) DISEASE DETECTION

Abstract

Hepatitis C (HCV) is a serious liver infection that can progress to cirrhosis or cancer, especially if early diagnosis is neglected. While conventional diagnostic methods such as ELISA are accurate, they are often limited in terms of efficiency and accessibility. This study introduces an innovative approach using a one-dimensional convolutional neural network (1D-CNN) for HCV disease classification, utilizing a Kaggle dataset consisting of 615 patient samples. The methodology includes data preprocessing such as handling missing values ​​and transforming categorical variables to numeric values ​​to ensure data readiness. The 1D-CNN model was trained using the Adam optimizer with ReLU activation functions in the convolution layer and sigmoid in the output layer. Model performance was comprehensively evaluated through accuracy, precision, recall, and F1-score metrics. The results showed that the 1D-CNN model achieved an accuracy of 83.74% on the training data and 81.30% on the testing data after hyperparameter tuning. This improvement is significant compared to the initial accuracy of only around 52%. However, the model exhibits a strong bias towards the majority class (Blood Donor), with very poor performance on minority classes such as Hepatitis, Fibrosis, and Cirrhosis. Nonetheless, this study contributes to the exploration of 1D-CNN for non-image medical data, which is still rarely studied. We conclude that despite the model's potential, further developments such as data balancing are needed to improve the model's overall generalization and accuracy.

Keywords: Deep Learning, Disease Classification, Hepatitis C, 1D CNN, Medical Diagnosis