Liver Diseases Prediction using KNN with Hyper Parameter Tuning Techniques
Sateesh Ambesange, Ranjana Nadagoudar, Rashmi Uppin, Vilaskumar Patil, Shruti Patil, Sushma Patil
Abstract
The healthcare industry is producing massive volumes of data. The algorithms using ML can be used to discover hidden patterns for making diagnosis and critical decisions. In the past years, Liver disorders have increased persistently and it is the reason for a significant number of deaths in many countries like India. The number of patients with liver disease are rapidly increasing due to several reasons like over consumption of alcohol, breathing in injurious gases, eating unhygienic food, pickles and drugs. The aim of our work is to predict liver disease by Machine learning based prediction model trained with the dataset fetched from the northeast of Andhra Pradesh, India. Feature analysis is performed on data and checked for balanced/imbalanced, distribution of data and correlation between the features and between features and liver disease. Transformation techniques have been used to transform data into normal distribution. So before performing transformation, outliers are detected and removed using outlier removal and the best features are selected based on correlation matrix and feature selection approaches, which will transform the data set effectively. Grid Search techniques are used in Hyper parameter tuning. Performance is evaluated using various metrics such precision, recall, f1-score, precision, precision-recall curve (PRC) and Receiver operating curve (ROC). Eight models are proposed out of which the fine tuning parameters of K-NN using Grid search gives the better performance of 91% accuracy.