UPSI Digital Repository (UDRep)
|
UPSI Digital Repository (UDRep)
|
|
|
|
||||||||||||||||||||||||||
| Abstract : Perpustakaan Tuanku Bainun |
| The increasing prevalence of ischemic stroke_particularly in Banjarmasin, Indonesia_demands the development of accurate, robust, and interpretable classification models to support timely and effective clinical decision-making. Conventional approaches and standard machine learning techniques often fall short when addressing the challenges posed by highly imbalanced medical datasets (91.40% majority vs. 8.60% minority) and limited model transparency, both of which impede clinical adoption. To overcome these limitations, this study introduces a rigorously optimized framework based on the XGBoost algorithm, enhanced by the Synthetic Minority Over-sampling Technique (SMOTE) to correct for class imbalance. The methodology incorporates a structured Train-Validation-Test split, 10-fold crossvalidation, and performance assessment using mean (_) and standard deviation (_). Two hyperparameter tuning strategies were implemented, with Random Forest employed as a comparative benchmark. SHapley Additive exPlanations (SHAP) were integrated to improve model interpretability. The XGBoost Hyperparameter Tuning Type 1 model, supported by Enhanced SMOTE, achieved a mean classification accuracy of 99.007% (_0.14%) and consistently exhibited high sensitivity (>97%) in detecting the minority class. Both ensemble models_XGBoost and Random Forest_ significantly outperformed the Decision Tree classifier, with no notable performance discrepancy between them. SHAP analysis consistently identified hypertension, heart disease, and genetic predisposition as key features contributing to classification outcomes. This research presents a robust and transparent machine learning framework for ischemic stroke classification, offering clinically relevant insights to aid in risk stratification and targeted intervention. The integration of SHAP enhances model explainability, thereby promoting greater trust among clinicians and informing improved strategies for stroke prevention and management in Banjarmasin. |
| This material may be protected under Copyright Act which governs the making of photocopies or reproductions of copyrighted materials. You may use the digitized material for private study, scholarship, or research. |