Integrating Causal Inference and Machine Learning for Early Diagnosis and Management of Diabetes

Abstract

In the context of the increasing prevalence of diabetes, this work focuses on integrating causal inference with Machine Learning (ML) for early diagnosis and effective management of diabetes. We applied a series of advanced techniques to improve model performance, including the use of data preprocessing methods, evaluation of variable importance and causal analysis, Feature Engineering methods, and hyperparameter optimization. The diabetes prediction model is a Stacking ensemble model that combines the predictions of several base models (namely: Random Forest Classifier, XGBClassifier, Gradient Boosting Classifier). Initial results showed a precision of 0.70, a recall of 0.70, an Area Under Curve (AUC) of 0.768, and a Mean Cross Entropy (MCE) of 0.299. After optimization, precision increased to 0.73, recall to 0.73, AUC to 0.798, and MCE improved to 0.271. This approach has demonstrated a significant improvement in diabetes prediction, suggesting that the integration of causal inference and Machine Learning is a promising path for the diagnosis and management of diabetes. The reduction in MCE, alongside improvements in precision, recall, and AUC, underscores the effectiveness of our optimization techniques in enhancing model reliability and performance.