Optimizing Liver disease prediction with Random Forest by various Data balancing Techniques

Liver disease is a prominent disease other than heart attack, which is taking a lot of lives. Since most of the time liver disease is detected at a later stage leading to death. Number of liver patients is increasing because of several reasons like over consumption of alcohol, breathing in injurious gas, consuming polluted water and so on which can affect health parameters. Using a machine learning prediction models, liver diseases can be predicted using those health parameters in early stages. In this work to build the machine-learning model, Indian Liver Patient Dataset (ILPD) hosted at UCI.edu 1 is used, which is based on Indian patient and Random Forest (RF) algorithm is used to predict the disease with different preprocessing techniques. Data set is checked for skewness, outliers and imbalance using univariate and bivariate analysis and then suitable algorithms used to remove outliers and various oversampling and under sampling techniques are used to balance the data. Further refinement of model is done through hyper parameter tuning using grid search and feature selection. The final model provides 100% accuracy and also good score across different metrics.