| _version_ | 1866902249614082048 |
|---|---|
| author | GOKUL, BESTHA Dhanush, Umakanth Shanumga Priya VG |
| author_facet | GOKUL, BESTHA Dhanush, Umakanth Shanumga Priya VG |
| contents | <h1>Diabetes Prediction Using Machine Learning</h1> <p><strong>Author:</strong> Bestha Gokul<br><strong>Affiliation:</strong> Department of Life Sciences, School of Sciences, Garden City University, Bengaluru, India<br><strong>Supervisor:</strong> Dr. Shanmuga Priya V G</p> <h2>Abstract</h2> <p>Diabetes mellitus is one of the most prevalent chronic diseases worldwide, requiring early detection and preventive strategies to reduce complications. This study applies machine learning techniques to predict diabetes risk using the 2015 Behavioral Risk Factor Surveillance System (BRFSS) dataset containing over 250,000 health records. Preprocessing steps included handling missing values, encoding categorical features, and standardizing variables. Four machine learning algorithms—Logistic Regression, K-Nearest Neighbors (KNN), Random Forest, and XGBoost—were implemented. Performance was evaluated using accuracy, confusion matrices, and classification reports. XGBoost achieved the highest predictive accuracy (85%), followed by Logistic Regression (83%), Random Forest (82%), and KNN (80%). Feature importance analysis revealed BMI, age, hypertension status, and physical activity as the strongest predictors. Results demonstrate the potential of machine learning in healthcare analytics, particularly for early detection and risk assessment of diabetes.</p> <h2>Keywords</h2> <p>Diabetes prediction, machine learning, Logistic Regression, Random Forest, XGBoost, healthcare analytics</p> <h2>Introduction</h2> <p>Diabetes is a chronic disease characterized by elevated blood glucose levels due to impaired insulin production or utilization. Long-term complications include cardiovascular disease, nephropathy, neuropathy, and retinopathy. Early diagnosis is crucial for prevention and management. With the availability of large-scale health data, machine learning has emerged as a powerful tool for predicting disease risk. This study aims to evaluate the performance of multiple supervised machine learning models in predicting diabetes risk using health and lifestyle indicators.</p> <h2>Materials and Methods</h2> <h3>Dataset</h3> <p>The dataset was sourced from Kaggle, originally derived from the Behavioral Risk Factor Surveillance System (BRFSS, 2015). The target variable <em>Diabetes_012</em> categorizes individuals into non-diabetic (0), pre-diabetic (1), and diabetic (2). Features included age, BMI, smoking status, physical activity, alcohol use, and general health.</p> <h3>Preprocessing</h3> <ul> <li> <p>Missing values handled via removal and imputation.</p> </li> <li> <p>Categorical variables encoded using label and one-hot encoding.</p> </li> <li> <p>Standardization applied to continuous variables.</p> </li> <li> <p>Dataset split into 70% training and 30% testing.</p> </li> </ul> <h3>Machine Learning Models</h3> <ol> <li> <p><strong>Logistic Regression</strong> – interpretable linear classifier.</p> </li> <li> <p><strong>K-Nearest Neighbors (KNN)</strong> – instance-based algorithm.</p> </li> <li> <p><strong>Random Forest</strong> – ensemble of decision trees with feature importance.</p> </li> <li> <p><strong>XGBoost</strong> – gradient boosting model optimized for accuracy.</p> </li> </ol> <h3>Evaluation Metrics</h3> <ul> <li> <p>Accuracy</p> </li> <li> <p>Confusion matrix</p> </li> <li> <p>Precision, recall, and F1-score</p> </li> </ul> <h2>Results</h2> <ul> <li> <p><strong>XGBoost:</strong> 85% accuracy (best performing model)</p> </li> <li> <p><strong>Logistic Regression:</strong> 83% accuracy</p> </li> <li> <p><strong>Random Forest:</strong> 82% accuracy</p> </li> <li> <p><strong>KNN:</strong> 80% accuracy</p> </li> </ul> <p>Feature importance analysis identified BMI, age, hypertension, and physical activity as key predictors. Confusion matrices and classification reports confirmed the robustness of XGBoost in handling class imbalance.</p> <h2>Discussion</h2> <p>Machine learning provides accurate, scalable, and cost-effective methods for diabetes prediction. XGBoost outperformed other models due to its boosting mechanism and regularization, while Logistic Regression offered clinical interpretability. Random Forest provided balanced performance and feature insights. KNN was less effective for large datasets.</p> <h2>Conclusion</h2> <p>This study demonstrates that machine learning, particularly XGBoost and Logistic Regression, can significantly enhance early diabetes risk prediction. Integration of such models into healthcare systems can support preventive strategies and clinical decision-making.</p> <h2>Future Work</h2> <ul> <li> <p>Incorporation of longitudinal data and electronic health records.</p> </li> <li> <p>Use of explainable AI (e.g., SHAP, LIME).</p> </li> <li> <p>Deployment of real-time mobile/web-based prediction tools.</p> </li> <li> <p>Fusion of structured data with wearable sensor data.</p> </li> </ul> <h2>References</h2> <p>(Selected from your project’s reference list – can be shortened for Zenodo)</p> <ol> <li> <p>Choudhury A., Gupta D. (2019). <em>Machine learning approach for the detection of diabetes.</em> Procedia Computer Science.</p> </li> <li> <p>Kavakiotis I. et al. (2017). <em>Machine Learning and Data Mining Methods in Diabetes Research</em> Computational and Structural Biotechnology Journal.</p> </li> <li> <p>Sisodia D., Sisodia D.S. (2018). <em>Prediction of diabetes using classification algorithms.</em> Procedia Computer Science.</p> </li> <li> <p>Islam M.M. et al. (2019). <em>Predicting diabetes using machine learning techniques: A comprehensive review.</em> Journal of Medical Systems.</p> </li> <li> <p>Zhou X., Qiao M., Liu Y. (2019). <em>An improved XGBoost model for diabetes prediction.</em> ICMLC Proceedings</p> </li> </ol> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_17144504 |
| institution | Zenodo |
| language | eng |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Diabetes Prediction Using Machine Learning GOKUL, BESTHA Dhanush, Umakanth Shanumga Priya VG Supervised Machine Learning <h1>Diabetes Prediction Using Machine Learning</h1> <p><strong>Author:</strong> Bestha Gokul<br><strong>Affiliation:</strong> Department of Life Sciences, School of Sciences, Garden City University, Bengaluru, India<br><strong>Supervisor:</strong> Dr. Shanmuga Priya V G</p> <h2>Abstract</h2> <p>Diabetes mellitus is one of the most prevalent chronic diseases worldwide, requiring early detection and preventive strategies to reduce complications. This study applies machine learning techniques to predict diabetes risk using the 2015 Behavioral Risk Factor Surveillance System (BRFSS) dataset containing over 250,000 health records. Preprocessing steps included handling missing values, encoding categorical features, and standardizing variables. Four machine learning algorithms—Logistic Regression, K-Nearest Neighbors (KNN), Random Forest, and XGBoost—were implemented. Performance was evaluated using accuracy, confusion matrices, and classification reports. XGBoost achieved the highest predictive accuracy (85%), followed by Logistic Regression (83%), Random Forest (82%), and KNN (80%). Feature importance analysis revealed BMI, age, hypertension status, and physical activity as the strongest predictors. Results demonstrate the potential of machine learning in healthcare analytics, particularly for early detection and risk assessment of diabetes.</p> <h2>Keywords</h2> <p>Diabetes prediction, machine learning, Logistic Regression, Random Forest, XGBoost, healthcare analytics</p> <h2>Introduction</h2> <p>Diabetes is a chronic disease characterized by elevated blood glucose levels due to impaired insulin production or utilization. Long-term complications include cardiovascular disease, nephropathy, neuropathy, and retinopathy. Early diagnosis is crucial for prevention and management. With the availability of large-scale health data, machine learning has emerged as a powerful tool for predicting disease risk. This study aims to evaluate the performance of multiple supervised machine learning models in predicting diabetes risk using health and lifestyle indicators.</p> <h2>Materials and Methods</h2> <h3>Dataset</h3> <p>The dataset was sourced from Kaggle, originally derived from the Behavioral Risk Factor Surveillance System (BRFSS, 2015). The target variable <em>Diabetes_012</em> categorizes individuals into non-diabetic (0), pre-diabetic (1), and diabetic (2). Features included age, BMI, smoking status, physical activity, alcohol use, and general health.</p> <h3>Preprocessing</h3> <ul> <li> <p>Missing values handled via removal and imputation.</p> </li> <li> <p>Categorical variables encoded using label and one-hot encoding.</p> </li> <li> <p>Standardization applied to continuous variables.</p> </li> <li> <p>Dataset split into 70% training and 30% testing.</p> </li> </ul> <h3>Machine Learning Models</h3> <ol> <li> <p><strong>Logistic Regression</strong> – interpretable linear classifier.</p> </li> <li> <p><strong>K-Nearest Neighbors (KNN)</strong> – instance-based algorithm.</p> </li> <li> <p><strong>Random Forest</strong> – ensemble of decision trees with feature importance.</p> </li> <li> <p><strong>XGBoost</strong> – gradient boosting model optimized for accuracy.</p> </li> </ol> <h3>Evaluation Metrics</h3> <ul> <li> <p>Accuracy</p> </li> <li> <p>Confusion matrix</p> </li> <li> <p>Precision, recall, and F1-score</p> </li> </ul> <h2>Results</h2> <ul> <li> <p><strong>XGBoost:</strong> 85% accuracy (best performing model)</p> </li> <li> <p><strong>Logistic Regression:</strong> 83% accuracy</p> </li> <li> <p><strong>Random Forest:</strong> 82% accuracy</p> </li> <li> <p><strong>KNN:</strong> 80% accuracy</p> </li> </ul> <p>Feature importance analysis identified BMI, age, hypertension, and physical activity as key predictors. Confusion matrices and classification reports confirmed the robustness of XGBoost in handling class imbalance.</p> <h2>Discussion</h2> <p>Machine learning provides accurate, scalable, and cost-effective methods for diabetes prediction. XGBoost outperformed other models due to its boosting mechanism and regularization, while Logistic Regression offered clinical interpretability. Random Forest provided balanced performance and feature insights. KNN was less effective for large datasets.</p> <h2>Conclusion</h2> <p>This study demonstrates that machine learning, particularly XGBoost and Logistic Regression, can significantly enhance early diabetes risk prediction. Integration of such models into healthcare systems can support preventive strategies and clinical decision-making.</p> <h2>Future Work</h2> <ul> <li> <p>Incorporation of longitudinal data and electronic health records.</p> </li> <li> <p>Use of explainable AI (e.g., SHAP, LIME).</p> </li> <li> <p>Deployment of real-time mobile/web-based prediction tools.</p> </li> <li> <p>Fusion of structured data with wearable sensor data.</p> </li> </ul> <h2>References</h2> <p>(Selected from your project’s reference list – can be shortened for Zenodo)</p> <ol> <li> <p>Choudhury A., Gupta D. (2019). <em>Machine learning approach for the detection of diabetes.</em> Procedia Computer Science.</p> </li> <li> <p>Kavakiotis I. et al. (2017). <em>Machine Learning and Data Mining Methods in Diabetes Research</em> Computational and Structural Biotechnology Journal.</p> </li> <li> <p>Sisodia D., Sisodia D.S. (2018). <em>Prediction of diabetes using classification algorithms.</em> Procedia Computer Science.</p> </li> <li> <p>Islam M.M. et al. (2019). <em>Predicting diabetes using machine learning techniques: A comprehensive review.</em> Journal of Medical Systems.</p> </li> <li> <p>Zhou X., Qiao M., Liu Y. (2019). <em>An improved XGBoost model for diabetes prediction.</em> ICMLC Proceedings</p> </li> </ol> |
| title | Diabetes Prediction Using Machine Learning |
| topic | Supervised Machine Learning |
| url | https://doi.org/10.5281/zenodo.17144504 |