A Genetic-based Framework for Diabetes Risk Prediction using Parental Genetic Contributions

The prediction of diabetes in children is gaining traction with respect to family genetic information-based early diagnosis. The application of advanced Machine Learning (ML) methodologies in the analysis of inherited traits and markers helps to develop a highly integrated prediction model based on the genetic influence of the parents. Five ML models-Naive Bayes, Random Forest, K-Nearest Neighbor (KNN), Support Vector Machine(SVM), and Logistic Regression are then tested based on accuracy and reliability over 5, 10, and 15 epochs in analyzing the potential models. The results indicate that the Random Forest model greatly outperformed other models with higher prediction accuracy. In order to avoid overfitting and to improve generalizability, feature selection, data preparation, and noise addition techniques are incorporated in the methodology. Important health and genetic characteristics of the children were identified as significant contributors. Confusion Matrix and Feature Correlation Analysis are ways to gauge performance, embellished by clear visualizations of feature distributions. This simple method shows promise in the integration of family factors with ML to early detect and accurately identify at-risk cases, thereby carrying the potential to predict, based on inherited information from parents, the diabetes risk in children accurately. To develop a model that assesses diabetes risk based on parental genetic information.