Investigating Gender and Age Variability in Diabetes Prediction: A Multi-Model Ensemble Learning Approach

Investigating Gender and Age Variability in Diabetes Prediction: A Multi-Model Ensemble Learning Approach

Diabetes affects over 500 million people globally, yet most prediction models treat all patients identically, ignoring critical demographic variability. This project delivers a data-driven study combined with a working software system that examines how prediction accuracy differs across gender and age groups. Using real-world diabetes datasets containing attributes such as age, gender, BMI, insulin levels, blood glucose, and lifestyle factors, the system trains and benchmarks multiple classifiers — Logistic Regression, K-Nearest Neighbors, Random Forest, XGBoost, and a multi-model ensemble — evaluating each through subgroup-wise accuracy, precision, recall, and ROC-AUC analysis. The project introduces fairness-aware evaluation metrics that expose demographic bias in standard models and proposes bias-mitigation strategies such as stratified resampling and calibrated ensembles. The final deliverable is an interactive diabetes risk prediction web application built in Python with visual analytics dashboards showing model performance per demographic segment. This makes it ideal for healthcare researchers, data science students, and institutions focused on fair machine learning in healthcare and responsible AI projects. The research findings and source code make this a high-impact final year project with strong publication potential.

Components

Python 3.8+
Scikit-learn
XGBoost
Logistic Regression / KNN / Random Forest
Pandas & NumPy
Plotly / Streamlit dashboards

Key Features

• Subgroup-wise accuracy, precision, recall and ROC-AUC analysis across gender and age segments
• Logistic Regression, KNN, Random Forest, XGBoost and multi-model ensembles benchmarked
• Fairness-aware evaluation metrics that expose demographic bias in standard models
• Bias mitigation using stratified resampling and calibrated ensembles
• Interactive diabetes risk prediction web app with per-demographic performance dashboards

Applications

Healthcare researchers, data science students and institutions focused on fair machine learning and responsible AI in healthcare. Strong publication potential.