Early Prediction of Diabetes using Machine Learning Classifiers

Abstract

Diabetes Mellitus is one of the most vastly dispersed, lethal and life threatening ailments not only around the globe but also in Bangladesh. It deteriorates the health condition gradually when the human body can not manufacture adequate insulin or could not acknowledge it in a decent fashion, which results in anomalously increased blood sugar levels. Countless complexities including high mortality, damages of numerous organs occur if the patients continue to live without medical treatment. So, identification of this illness in the premature phase and timely medical therapy can retain more humankind from serious injuries. The astonishing advancements in health sciences have contributed to a noteworthy volume of data. Machine learning algorithms have extensively gained popularity in medical science to diagnose and predict the likelihood of this sickness using these tons of raw data. The intention of this research work is to make a side by side analysis of multiple machine learning classifiers and their results of prognosis to this deadly disease beforehand. Decision Tree, Logistic Regression, Random Forest, Support Vector Machine, K-Nearest Neighbours and Naive Bayes have been applied in supervised circumstances to predict the possibility of the disease. The fresh dataset at hand is imbalanced and has been accumulated from UCI repository and having sixteen dimensions and one outcome class. That’s why pre-processing tasks like missing or null value replacement, label encoding, importance feature selection, SMOTE resampling methodology to balance class variables, have been conducted on the data. Scikit Learn, a python free module has been used for analysing and visualizing the experiments. Lastly, outcomes of the algorithms have been compared to put a verdict that the Random Forest classifier outperforms others with 98.38% accuracy level.