Analysis of Machine Learning and Deep Learning for Diabetes Diagnosis in Design Thinking

Diabetes is less likely to cause serious complications if it is diagnosed and treated early. The goal of the proposed study is to create a machine learning pipeline that produces the smallest possible feature subset with the highest accuracy for predicting the start of diabetes. In contrast to the well-known Pima Indians Diabetes Dataset (PIDD), it uses a new diabetes dataset that is sufficiently representative and gender-neutral. Health professionals from all across the world have expressed serious worries concerning the need to treat and identify it as early as possible. The most effective method for diagnosing diabetes disease is now made possible by advanced technological and data mining methods.In order to detect diabetes based on various symptoms and traits, this review paper compares several data mining techniques and studies a classifier. In the machine learning, several feature selection methods are used towardcreate a reduced feature subset with the purpose is fed addicted todiverse heterogeneous classifiers. Feature selection is used to identify the most important features from dataset. On the preprocessed dataset, classification is done by using a variety of heterogeneous classifiers like Naive Bayes (NB), Logistic Regression (LR), K-Nearest Neighbor (KNN), Decision Tree (DT), Support Vector Machine (SVM), Random Forest (RF), AdaBoost, Deep Learning, and Gradient Boost (GB) as base learners followed by stacking ensemble. Precision, recall, and F1-Score metrics are used to assess each classifiers performance, as well as the performance of deep learning classifiers. Deep learning outperforms the other classifiers, according to a comparison with more recent research in the same field on the same dataset.