A Robust Multivariate Logistic Regression Model for Smart Parking Occupancy Prediction

Purpose: This study developed an enhanced multivariate logistic regression (MLR) model integrated with robust ordinary least squares (ROLS) techniques to address parking occupancy prediction challenges in rapidly urbanizing environments. Focusing on developing country contexts with infrastructure constraints, the research targeted three limitations of conventional approaches: vulnerability to data anomalies, insufficient interpretability, and poor adaptation to resource-limited settings. Methodology: Employing design science research (DSR) methodology, the study utilized parking datasets from Kaggle and GitHub repositories. Comprehensive preprocessing included ROLS-based outlier treatment and temporal/environmental feature engineering. The model incorporated SHapley Additive exPlanations (SHAP) for interpretability and underwent hyperparameter optimization via grid search. Evaluation employed an 80-20 train-test split with accuracy, precision, recall, F1-score, and AUC-ROC metrics. Findings: The ensemble model achieved superior performance (R²=0.9007, MSE=0.00878, accuracy=91.56%) compared to standalone MLR (84.31% accuracy) and ROLS (MSE=0.00872) implementations. Key predictors included historical occupancy patterns, temporal variables, and weather conditions. SHAP analysis confirmed the model's operational transparency while maintaining computational efficiency. Unique Contribution to Theory, Practice and Policy: Implementation in real-time smart parking systems through IoT networks is recommended. Future research should pursue: 1) cross-regional validation studies, 2) dynamic pricing algorithm integration, and 3) enhanced anomaly detection mechanisms. The study provides a theoretically grounded yet practical solution optimized for developing urban contexts.