Classification of Diabetes using Multinomial Naive Bayes, Logistic Regression, and Multi-Layer Perceptron Algorithms

Diabetes is an ongoing condition in which a human being's blood glucose levels increase to unacceptably high levels. For the purpose of organizing the required treatments and avoiding the development of more severe diseases that this disease may bring on, diabetes should be detected as early as possible. In this study, diabetes is classified through the use of different models and to determine the most appropriate model that can be used for this problem. In this study, Logistic Regression, Multinomial Naive Bayes, And Multi-Layer Perceptron Algorithms are utilised as classification models. The Indian Pima Data Set is utilised to test these techniques. The preprocessing steps used in this study involve working with the noisy data, scaling of data using normalization, processing imbalanced data using the SMOTE approach, and using sequential backward selection technique (SBS) for features selection. The classification performances of techniques Logistic Regression, Multinomial Naive Bayes, and Multi-Layer Perceptron obtained by dividing dataset into 80% training dataset and 20% testing dataset are 74.5%, 78%, and 62%, respectively. This study has specifically solved the issues of under fitting and overfitting.