Enhancing Non-Alcoholic Fatty Liver Disease Prediction with Machine Learning and Recursive Feature Elimination

Non-alcoholic disease detection is one of the leading research works in recent days. Modern life has changed the food and environmental culture, making them overweight, stressed, unhealthy conditions always and which causes various diseases due to overweight and diabetes. Commonly, an alcoholic addict can be affected by Fatty Liver Diseases (FLD), whereas identifying fatty liver diseases for a non-alcoholic person is a challenging task. It is not so easy even suspecting that a patient has FLD at the earlier stage of the symptoms since the symptoms of FLD are very similar to other diseases, and it may lead to wrong diagnosis and treatment. The severity level of 30% of FLD patients is increased suddenly and leads to heart attack, stroke, and death. Thus, based on the symptoms of weight loss, abdominal pain, and fatigue, it is essential to diagnose NAFLD, which can be identified accurately from pathological and genomic data using efficient learning methods to provide the right and better treatment immediately. This paper implements multiple machine learning algorithms for analyzing the pathological information obtained from the NAFLD and NASH DNA datasets and finding the best model concerning the performance. This paper uses 3-fold cross verification with recursive feature elimination methods to improve the original accuracy of the prediction. The performance comparison shows that the SVM model obtained 87% accuracy, which is better than the KNN and RF models. The experimental results with the performance comparison are explained in detail in the paper.