A Comparative Study of Feature Selection Methods for EHR Diagnosis Codes in Opioid Use Disorder Prediction

Selective prediction as a triage gate for primary-care depression screening: quantifying and mitigating selection bias in CHARLS-2011

Background Primary care in China lacks structured mental-health assessment, and the machine-learning models that could support such screening are typically developed on heavily selected samples. Cumulative inclusion and exclusion criteria, though usually treated as neutral data-cleaning steps, can create heterogeneity in predictive reliability among retained participants. Using the China Health and Retirement Longitudinal Study (CHARLS) 2011 baseline, we quantified how selection funnels distort epidemiological associations and inflate machine-learning metrics, and tested selective prediction as mitigation. Methods Using the CHARLS 2011 baseline with temporal external validation in CHARLS-2018, we built a four-level selection funnel (L0-L3), evaluated five classifiers with nested cross-validation and SMOTE, and compared model-embedded uncertainty with a decoupled predictor-selector framework; XGBoost cross-validation residuals drove risk stratification and classification and regression tree (CART) rules. Results Sample sizes fell from L0 n=17,705 to L3 n=4,256 (24.0%). The cancer-depression odds ratio attenuated from 1.78 (95% CI 1.32-2.41) to 1.39 (0.74-2.63), losing significance. AUC rose with selection but not after multiple-comparison correction, whereas calibration error increased for four of five models. Model-embedded uncertainty succeeded only for XGBoost; with the decoupled XGBoost residual selector, all five models achieved selective prediction at approximately 20% coverage (test AUC 0.90, 95% CI 0.85-0.95), abstaining on approximately 80% of cases for individual safety. Risk stratification was stable (residual Spearman correlations >0.95; multi-seed Jaccard 0.88), and CART rules used self-rated health, education, pain, and marital status. Conclusions The findings support a deployable primary-care triage pathway: a four-variable rule identifies patients suitable for algorithm-assisted scoring (approximately 20% coverage) and routes the remainder to human evaluation. Methodologically, cumulative selection bias produces a dual distortion: epidemiological associations are compressed and machine-learning metrics inflated. Selective prediction is limited mainly by uncertainty-indicator design. Performance metrics should be reported with selection level, coverage, and calibration trajectory. Decoupled selective prediction with CART rule extraction provides an actionable framework for quality-controlled, tiered-care deployment. Keywords: selective prediction, selection bias, CHARLS, depression, predictor-selector decoupling, uncertainty quantification, classification and regression tree, triage, clinical decision support, health management.

An interpretable machine learning framework for early-stage diabetes mellitus prediction using comparative classification models and SHAP

A machine learning-based framework enhanced with explainability is introduced, built around a structured data preparation process that handles categorical encoding, numerical scaling, and minority class oversampling through the SMOTE technique, positioning it as a trustworthy tool for assisting medical professionals in data-driven clinical decision-making.

Data-Driven Stroke Prediction: Leveraging KNN for Early Detection in Clinical Datasets

This study explores the application of the K-Nearest Neighbors (KNN) algorithm in predicting stroke risk using a healthcare dataset from Kaggle containing 5,110 patient records with demographic and clinical attributes. Preprocessing steps included categorical encoding, normalization, and feature engineering, followed by model training using a pipeline that incorporated SMOTE to address class imbalance. Hyperparameter tuning via GridSearchCV improved model performance. The final KNN model achieved an accuracy of 84%, with a recall of 0.64 and precision of 0.18 for the stroke class, indicating moderate sensitivity but significant false positive rates typical of imbalanced medical datasets. Evaluation metrics included the confusion matrix, ROC curve (AUC = 0.82), and precision-recall curve (AP = 0.19). While the model demonstrates reasonable discriminative ability, the low precision suggests limited utility as a standalone diagnostic tool. These findings contribute to the growing body of research on machine learning for clinical decision support, emphasizing the need for balanced performance metrics in imbalanced datasets and highlighting KNN's potential as a preliminary screening instrument when combined with clinical judgment.

Data diversity vs. model complexity in the prediction of pediatric bipolar disorder: Evidence from academic and community clinical samples.

Pediatric bipolar disorder is challenging to diagnose accurately due to symptom heterogeneity. More standardized and data-driven approaches are needed to enhance diagnostic reliability. We evaluated a clinical decision tool (nomogram), statistical methods (logistic regression, LASSO), machine learning (support vector machine, random forest, k-nearest neighbors, extreme gradient boosting), and deep learning (multilayer perceptron) for pediatric bipolar disorder prediction across two datasets collected in academic (N = 550) and community (N = 511) clinical settings. We compared three modeling strategies: cross-dataset validation, cross-dataset with interaction terms, and pooled-dataset. We assessed model performance using discrimination, calibration, and predictor importance ranking. In the baseline cross-dataset approach, all models showed good internal discrimination in the academic dataset, but external discrimination in the community dataset substantially declined. Interaction-enhanced models slightly improved internal discrimination but not external performance or calibration. Recalibration substantially improved cross-dataset calibration. Models trained on the pooled sample showed strong performance on held-out samples drawn from the heterogeneous pooled cohort, with good calibration for most models. Across models and training strategies, PGBI-10M was consistently identified as the most important predictor. Predictive models for pediatric bipolar disorder showed strong internal performance but limited cross-setting generalizability due to dataset shift and miscalibration. Within the present study, increasing model complexity did not improve external performance, whereas training on pooled data improved performance on held-out samples from the heterogeneous pooled cohort. These findings suggest that training-data diversity may provide greater practical benefit than increasing model complexity for developing robust psychiatric prediction models, underscoring the importance of open and collaborative datasets.

Relevance Based Prediction: A Transparent, Non-Artificial Intelligence, Mathematical Solution to Personalized Opioid Treatment

Feature Correlation and Class Balancing Effects on Random Forest Accuracy in Diabetes Risk Prediction Using the Teboul Health Indicators Dataset

Early detection of diabetes is essential for reducing disease burden and long-term healthcare costs. This study presents a Random Forest-based predictive framework for diabetes risk assessment using the Teboul Health Indicators Dataset, which is derived from the Centers for Disease Control and Prevention (CDC) Behavioral Risk Factor Surveillance System (BRFSS 2015). The proposed framework incorporates feature correlation analysis and evaluates the effects of four class-balancing strategies, namely under-sampling, oversampling, Synthetic Minority Oversampling Technique (SMOTE), and class weighting, on classification performance. The dataset, comprising 253,680 records and 21 health-related attributes, was preprocessed using normalization and stratified random sampling. Correlation analysis identified General Health (GenHlth), High Blood Pressure (HighBP), Body Mass Index (BMI), and Age as the most influential predictors of diabetes risk. Experimental results demonstrated that the under-sampling approach achieved the best trade-off between precision and sensitivity, yielding an F1-score of 0.426, a recall of 0.777, and an area under the receiver operating characteristic curve (AUC) of 0.808, outperforming the other balancing methods. The findings indicate that appropriate class balancing improves sensitivity and fairness in diabetes classification, while Random Forest provides model interpretability through feature-importance analysis. Future work will focus on integrating Explainable Artificial Intelligence (XAI) techniques and hybrid ensemble balancing methods to further enhance transparency and predictive performance in healthcare analytics.