An AI-based Liver Disease Prediction Model based on Pearson Correlation Feature Selection Method

Liver disease is a very critical disease in today's world. There are various types of liver disorders, including some that are brought on by viruses. Detecting liver infections in their early stages is crucial for more effective treatment. An AI-based automated diagnostic model can play a very significant role in detecting liver illness. The main goal of this research is to create an AI based hybrid model utilizing feature selection and classification algorithms to detect liver disease. Three feature selection techniques - Pearson Correlation, Feature Importance using Extra Tree, and Mutual Information Gain are used on the ILPD dataset to identify the relevant features. The Decision Tree (DT), K-Nearest Neighbour (KNN), Random Forest (RF), Adaptive Boosting (Adaboost), and Extreme Gradient Boosting (XGboost) classifiers have been used with the selected features of the dataset. The performance of models has been evaluated with various performance parameters, namely accuracy, precision, sensitivity, and F-Measure. The combination of the Pearson Correlation algorithm with the Random Forest classifier has shown superior performance compared to other classifiers like DT, KNN, RF, Adaboost, and XGBoost. The finding also depicts that Pearson Correlation algorithm have effectively eliminated irrelevant features from the data set, and the feature selection ratio of Pearson Correlation Algorithm is 80%. This proposed PC-RF model has provided 80% accuracy in identifying liver illness, which is 3% to 8% better accuracy than the other classifiers such as DT, KNN, Adaboost, and XGboost. Additionally, the proposed PC-RF model has achieved 4% to 25% better accuracy over latest state-of-the-art models.