Performance of machine learning approaches on prediction of esophageal varices for Egyptian chronic hepatitis C patients

Esophageal Varices is one of the most common side-effects of liver cirrhosis diseases which is detected by Upper endoscopy. Screening all patients implies many endoscopies will be needed, which increases the workload of endoscopy units. The aim of this study is to find solutions to diagnose the disease, by analyzing the patterns found in the data through classification analysis, using machine learning techniques for early prediction in cirrhotic patients based on their clinical examination. This research study attempts to propose a quicker and more efficient technique for disease diagnosis, leading to timely patient treatment. Our method analyzed 4962 patients with chronic hepatitis C from fifteen different centers in Egypt between 2006 and 2017. The dataset included twenty-four individual clinical laboratory variables. Esophageal Varices was present in 2218 patients and absent in 2, 744 patients. Different types of feature selection (Filter-Wrapper) Approaches were applied to select the most significant features. The proposed model used six common algorithms including Neural Networks, Naïve Bayes, Decision Tree, Support Vector Machine, Random Forest and Bayesian Network to achieve our objective. The results showed that correlation and (p-value) based on filter method and Bayesian Network algorithm are well-suited for this analysis. Only nine variables: Gender, Platelet, Albumin, Total Bilirubin, BaselinePCR, Liver, Spleen, Stiffness, and prothrombin concentration were the most significant predictors for Esophageal Varices. The Bayesian Network algorithm showed the highest performance; it achieved 74. 8% and 68. 9% for Area Under Receiver Operating Characteristic curves and accuracy, respectively. To conclude, machine learning techniques were able to predict Esophageal Varices in cirrhotic patients. The experimental results show that the Bayesian Network achieved better results than the other approaches.