Izvorni znanstveni članak
https://doi.org/10.7906/indecs.24.6.2
Transforming Academic Library Workflows with Artificial Intelligence: Bibliometric Analysis and Mini-Review
Milica Tufegdžić
; Academy of Proffesional Studies, Information Technology Department, Kragujevac, Serbia
Đorđe Mirjanić
; University of Banja Luka, Faculty of Medicine, Department of Stomatology, Banja Luka, Bosnia and Herzegovina
Predrag Dašić
; Engineering Academy of Serbia, Belgrade, Serbia
*
Nevena Tufegdžić
; SaTCIP Publisher Ltd., Vrnjačka Banja, Serbia
* Dopisni autor.
Sažetak
This study evaluated regression and classification models for predicting Artificial Intelligence diagnostic confidence and patient satisfaction in healthcare using a synthetic dataset of 5 000 patients with 23 variables generated through clinically informed rules. Linear Regression, Support Vector Regression, Random Forest, HistGradientBoosting, eXtreme Gradient Boosting, Light Gradient Boosting Machine, and CatBoost were tested for regression to estimate the target variable AI_Diagnosis_Confidence. For multiclass classification of Patient_Satisfaction_Class (Low, Medium, High), models evaluated were Logistic Regression, Support Vector Machine, RF, and HGBoost. Model performance was assessed using cross-validation with a coefficient of determination R², mean absolute error, and root mean squared error for regression, and accuracy, balanced accuracy, F1 macro, precision macro, and recall macro for classification. SHapley Additive exPlanations (SHAP) was used to interpret predictions and quantify feature contributions, with Linear Regression achieving the best regression performance (R² = 0.9577) with Diagnosis Complexity, Comorbidity Count, and Lab Abnormality Score as the most influential predictors. For classification, LogR delivered the most balanced performance, capturing minority classes, while tree-based models were biased toward the majority class. SHAP highlighted Recovery Time, Diagnosis Complexity, and Total Risk Score as influential features guiding class-specific predictions. These findings show that interpretable models with SHAP can accurately and transparently predict Artificial Intelligence diagnostic confidence and patient satisfaction, with simpler models excelling when relationships are linear.
Ključne riječi
patient satisfaction; predictive modelling; regression; classification; interpretable artificial intelligence
Hrčak ID:
350255
URI
Datum izdavanja:
30.12.2026.
Posjeta: 0 *