AI-DRIVEN SUSTAINABLE URBAN PARKING FRAMEWORK FOR GREEN COMPUTING AND SMART CITY OPTIMIZATION

Urban parking management is a growing challenge in smart cities, where data-driven decision-making can significantly improve enforcement and resource optimization. Traditional manual screening methods face significant issues, including human error in ticketing and categorization, delays in processing due to paperwork and field inspections, and lack of predictive analytics to anticipate high-risk zones. These limitations hinder scalability, consistency, and real-time responsiveness in densely populated urban areas. To overcome these issues, a Djangobased web application is proposed, integrating automated data preprocessing, secure rolebased access, and model deployment tailored to urban parking violations. The system allows users to upload raw CSV datasets, which are automatically cleaned using IQR-based Winsorization, label encoding, and missing value imputation. The preprocessed data is split and passed through several models, with trained models saved for reuse using Joblib. Among these, the CatBoost Classifier is highlighted for its excellent handling of categorical data and reduced overfitting, making it ideal for this use case. It classifies new test records into Idle_Parking or Overflow_Parking, and displays results through confusion matrices and classification reports. In this study, four machine learning models—Decision Tree, Random Forest, Logistic Regression, and CatBoost—were evaluated on urban parking violation datasets. The CatBoost Classifier achieved the highest performance with an accuracy of 99.98%, precision of 99.98%, recall of 99.98%, and F1- score of 99.98%, outperforming Random Forest (99.87%), Decision Tree (99.64%), and Logistic Regression (97.92%). Confusion matrix analysis confirmed that CatBoost misclassified only 9 total instances, compared to 74 (Random Forest), 1138 (Logistic Regression), and 22,766 (Decision Tree).