Hybrid Machine Learning Algorithms for Early Prediction of Diabetes

With changes in people’s lifestyles, several health-related issues are occurring, leading to severe damage. Diabetes is one such disease. It is one of the fastest-growing diseases in today’s world, affecting lakhs of people every year. Diabetes is mainly caused by a lack of secretion of insulin in the blood, which leads to the accumulation of glucose in the bloodstream. This increases the glucose level in the blood, affecting other organs of the body, such as the heart, nerves, and kidneys. An early diagnosis will help the patient better deal with the condition. About 537 million people (ages 20 to 79) worldwide had diabetes in 2021. By 2030, the number of people with diabetes is expected to increase to 643 million, and by 2045, it is expected to rise to 783 million. This exponential growth in projections requires serious attention. The present chapter focuses on the use of various machine-learning algorithms for the early prediction and detection of diabetes. After a thorough literature review, six algorithms were selected for diabetes prediction. To compare the performance of different machine learning algorithms, three different datasets were used. The models are then measured based on the confusion matrix, accuracy, support, precision, and recall measures. The comparative analysis of these models across two different datasets showed that random forest has high accuracy, and logistic regression has the highest precision. To further improve the performance of the classifiers, several preprocessing techniques are employed in the present work, which include principal component analysis (for reducing the number of attributes), random oversampling (for balancing the dataset), one-hot encoding (for categorical data), standard scaling (for standardization), and stratified k-fold cross-validation (for resampling). The experimental results show that the performance of all the classifiers has improved after the application of different preprocessing techniques across all performance measures used in the study for both datasets.