Machine Learning Approaches to Identify Influential Factors of the Comorbid Psychiatric Symptoms and Hypertension Among Rural Adults in Bangladesh: A Cross‐Sectional Study

Figures

• 

  Figure 1. Schematic diagram of the sample section.

• 

  Figure 2. Schematic presentation of selecting the best machine learning model. ML, machine learning.

• 

  Figure 3. ROC‐AUC, PR and calibration curve. •ROC‐AUC, receiver operating characteristics—area under the curve; PR, Precision‐recall. •GB, Gradient Booster; KNN, K Nearest Neighbor; LR, Logistic Regression; RF: Random Forest; SVM, Support Vector Mechine; XGB, Extreme Gradient Booster. •Higher values for ROC and PR indicate a better model, and •all models showed lower PR scores. This result is consistent with the low prevalence of the outcome in the dataset (139/1603), which makes PR metrics particularly sensitive to class imbalance. The proximity of the curves to the baseline indicates that the models exhibit only modest differentiation of the positive class, with performance broadly aligned with random prediction. •The calibration plot shows the agreement between predicted probabilities and observed outcomes; curves closer to the diagonal 45‐degree line indicate better calibration. In this study, the SVM model demonstrated comparatively better calibration than other models.

• 

  Figure 4. SHapley Additive exPlanations (SHAP) Beeswarm plot shows variables that explain comorbid psychiatric symptoms and hypertension in the Support Vector Machine model. The SHAP plot provides an overview of variable influence, showing the direction of each variable's impact through the distribution of red and blue dots. The vertical bar on the right represents the range of values for the independent variables, with blue indicating low values and red indicating high values. A positive SHAP value of a variable on the x‐axis corresponds to a higher likelihood of the outcome, while a negative value suggests a lower likelihood, such as no comorbid conditions. For instance, the variable “CVD” is coded as 0 for no and 1 for CVD. A value of 0 corresponds to the blue color, and 1 corresponds to red on the bar. CVD emerged as a significant risk factor of comorbid psychiatric symptoms and hypertension, as the red dots shifted towards positive SHAP values, indicating that individuals who had CVD were more likely to develop comorbidity.

• 

  Figure 5. Assessing the effect size of influential factors in predicting the comorbid psychiatric symptoms and hypertension. * indicates the name of the variables; BMI, body mass index.

• 

  Figure 6. SHapley Additive exPlanations (SHAP) dependency plot explaining the interaction effects of variables explaining the comorbid psychiatric symptoms and hypertension in the Support Vector Machine model.