Abstract
Diabetes being a metabolic disease is affecting people worldwide. It brings them under the life risk often. Early diagnosis and prognosis are required to control the disease's associated health problems and complications to prevent the damage of internal organs that may become fatal to life. In this research efficient deep learning models naming: XGBOOST and LGBM have been implemented on the dataset provided during WiDS 2021 Datathone. Before applying deep learning algorithms, an extensive feature engineering process was administered to get better insight into correlated features. It highlighted the concerns including filling missing values, class imbalanced, and age groups as important participatory factors in predicting anomalous results. Data segregation on groups including gender, age, ethnicity and max glucose didn't show visible differences among the classes so the machine learning procedures were implemented on the complete dataset. Finally, model evaluation was carried out using ROC evaluation method with 0.87 accuracy.