Early-stage diabetes prediction using machine learning algorithms

Abstract

Data mining has gained a significant amount of emphasis over the past few decades for its ability to retrieve insights from big datasets in order to identify trends and develop links that may be used to address issues. There are many machine learning (ML) algorithms being utilised for predictive analysis based on large datasets. The ability of ML algorithms to assist in predictive analysis can help in diagnosing diseases at the right time and prove as a great help to the medical field. One such illness is 'Diabetes'. Since diabetes is a widespread condition with weak early signs, people will benefit from a reliable prediction tool to aid in self-diagnostics. This will also assist in early identification of diabetes to prevent it from reaching a complex stage. This study focuses on utilising the popular ML algorithms namely, SMO, Multilayer Perceptron, AdaBoost, Random Forest and Bagging using WEKA tool, to build models to correctly predict diabetics' disease at an early stage based on symptoms to take corrective measures in lifestyle. The dataset utilised for this study was taken from the Sylhet Diabetes obtained from Kaggle. After evaluation of these algorithms, it was concluded that Random Forest provides better results compared to the other algorithms by providing the accuracy of 94.8718 %.