Diabetes Prediction using Feature Extraction and Machine Learning Models

Diabetes is a disorder wherein the production of insulin is stopped or it is not responded to by the body. This results in high blood glucose. Our food produces blood glucose and insulin produced by the pancreas makes it possible for the glucose to reach the cells. High blood sugar may lead to kidney failure or heart disease. The risk and the severity of diabetes can be reduced by predicting it at an early stage. In this research paper, we strive to propose machine learning algorithms namely KNN, SVM, Decision Tree, Random Forest, LGBM, and Adaboost that help predict diabetes. All these algorithms are built using the Pima Indian Diabetes (PID) dataset by taking several parameters such as glucose, skin thickness, insulin, age, etc. that help locate diabetes. We trained our models and observed that LGBM outperformed all other models by giving an accuracy of 89.85% and AUC as 0.95. Thus, LGBM is more efficient algorithm in distinguishing diabetic and non-diabetic person.