Scholarlink

Background <p>Depression co-occurring with non-communicable diseases among older adults represents a significant public health concern in India, particularly in the context of rapid population ageing. The growing availability of large-scale, nationally representative dataset has created opportunities for data-driven prediction of mental health outcomes. However, limited comparative evidence is available on the performance of traditional statistical methods versus modern machine learning techniques using large datasets. In the present era, a comparative evaluation of traditional statistical models and advanced machine learning techniques is needed to support more accurate and efficient prediction approaches in clinical and health care field.</p> Methods <p>This study utilized data from Wave 1 of the Longitudinal Ageing Study in India (LASI, 2017–2018) and included 59,298 older adults aged 45&#xa0;years and older. The outcome variable is major depressive symptoms, was assessed using the CES-D-10 scale (cut-off ≥ 4). Traditional logistic regression was compared with seven ML algorithms i.e., Random Forest, Decision Tree, Naive Bayes, K-Nearest Neighbors, Support Vector Machine, Neural Network and Ridge Classifier. All models were trained and evaluated using stratified 10-fold cross-validation. Performance was assessed using accuracy, F1-score, AUROC, RMSE, MAE, precision and recall.</p> Results <p>Random Forest achieved the strongest performance with an accuracy of 97.4%, F1-score of 0.974 and AUROC of 0.998. Decision Tree and KNN also outperformed traditional logistic regression that showed an accuracy of 60.5% and AUROC of 0.641. ML models were notably more effective at capturing nonlinear patterns and high-dimensional interactions. However, the exceptionally high performance of ensemble models suggests the need for future validation to rule out potential overfitting or data leakage.</p> Conclusion <p>Ensemble-based ML models demonstrated superior predictive ability compared to logistic regression for identifying older adults at risk of depressive symptoms. These findings highlight the potential of integrating ML-driven decision-support tools into mental health screening and NCDs management efforts in India. Further research should include external validation and interpretability analyses to enhance clinical applicability.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Artificial intelligence based prediction of depression among older adults with NCDs in India using traditional logistic regression and machine learning models

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月19日

查看全文
  • Kanchan Yadav,
  • Dechenla Tshering Bhutia

摘要

Background

Depression co-occurring with non-communicable diseases among older adults represents a significant public health concern in India, particularly in the context of rapid population ageing. The growing availability of large-scale, nationally representative dataset has created opportunities for data-driven prediction of mental health outcomes. However, limited comparative evidence is available on the performance of traditional statistical methods versus modern machine learning techniques using large datasets. In the present era, a comparative evaluation of traditional statistical models and advanced machine learning techniques is needed to support more accurate and efficient prediction approaches in clinical and health care field.

Methods

This study utilized data from Wave 1 of the Longitudinal Ageing Study in India (LASI, 2017–2018) and included 59,298 older adults aged 45 years and older. The outcome variable is major depressive symptoms, was assessed using the CES-D-10 scale (cut-off ≥ 4). Traditional logistic regression was compared with seven ML algorithms i.e., Random Forest, Decision Tree, Naive Bayes, K-Nearest Neighbors, Support Vector Machine, Neural Network and Ridge Classifier. All models were trained and evaluated using stratified 10-fold cross-validation. Performance was assessed using accuracy, F1-score, AUROC, RMSE, MAE, precision and recall.

Results

Random Forest achieved the strongest performance with an accuracy of 97.4%, F1-score of 0.974 and AUROC of 0.998. Decision Tree and KNN also outperformed traditional logistic regression that showed an accuracy of 60.5% and AUROC of 0.641. ML models were notably more effective at capturing nonlinear patterns and high-dimensional interactions. However, the exceptionally high performance of ensemble models suggests the need for future validation to rule out potential overfitting or data leakage.

Conclusion

Ensemble-based ML models demonstrated superior predictive ability compared to logistic regression for identifying older adults at risk of depressive symptoms. These findings highlight the potential of integrating ML-driven decision-support tools into mental health screening and NCDs management efforts in India. Further research should include external validation and interpretability analyses to enhance clinical applicability.