Efficient Feature Selection for Prediction of Diabetic Using LASSO

Abstract

Machine learning systems provide the most accurate predictions using the data and the automatic-learning mechanism. Performance of the machine learning systems prediction fully depends on the efficient selection of features. Feature selection is one of the critical and challenging tasks in statistical modeling and machine learning. The main aim of this paper is to present a novel method for selecting efficient features for predicting diabetes using the Least Absolute Shrinkage and Selection Operator (LASSO) method. LASSO is one of the key models that works efficiently for any kind of problem in the purpose of feature selection by minimizing the prediction error. Diabetes is becoming an alarming common disease in the world, including Sri Lanka. The discovery of knowledge from diabetes datasets is important in order to make an effective diabetes diagnosis. The study dataset here we used, based on the Sri Lankan people and the identified feature variables used for the analysis are, age, gender, height, weight, BMI value, waist circumference, fasting blood sugar level, postprandial blood sugar level, HbA1C level, total cholesterol level, high-density lipoproteins level, triglycerides level, number of physical activity hours and the number of sedentary life hours. Results of the study confirm the efficiency and effectiveness of the proposed approach for the selection of the most significant features of diabetic data. This study will help to build a model using the selected features that can predict diabetes using machine learning systems.