Behavioural Risk Factor Surveillance of Diabetes Using Supervised Learning

The CDC conducts the Behavioral Risk Factor Surveillance System (BRFSS) annually, a health-related telephone survey, gathering responses from more than 400,000 Americans since 1984. It focuses on health-related risk behaviors, chronic conditions, and the use of preventive services. In study, we utilized a CSV dataset from Kaggle for the year 2015, containing 253,680 survey responses to the BRFSS2015. The dataset comprises 21 feature variables derived from questions posed to participants or calculated from their responses. The target variable, Diabetes binary, categorizes individuals as having either no diabetes (0) or prediabetes/diabetes (1). The dataset is imbalanced, reflecting real-world distributions of health conditions. We utilized a range of machine learning algorithms, including decision trees, logistic regression, gradient boosting machine, random forest, k-nearest neighbors, support vector machines, XGBoost, Naive Bayes, feed-forward neural networks, and Ada Boost, to predict diabetes status based on the provided features. Evaluation metrics like accuracy, F1-score, and confusion matrices were employed to assess model performance across different algorithms. Through this analysis, our goal is to offer insights into the effectiveness of different machine learning techniques in predicting diabetes and highlight the significance of feature selection and model evaluation in healthcare applications.