A Comparative Analysis of Machine/Deep Learning Models for Parking Space Availability Prediction.

Area of Science:

• Computer Science
• Artificial Intelligence
• Data Science

Background:

• Machine/Deep Learning (ML/DL) models are increasingly used for data analysis and prediction.
• Algorithm performance varies significantly based on data characteristics and application domain.
• Identifying the most suitable ML/DL algorithm for a specific task is crucial for optimal outcomes.

Purpose of the Study:

• To conduct a comparative analysis of various ML/DL algorithms for predicting parking space availability.
• To determine the most effective algorithm for parking prediction using a real-world dataset.

Main Methods:

• Evaluated Multilayer Perceptron, K-Nearest Neighbors, Decision Tree, Random Forest, and Voting Classifier algorithms.
• Utilized Santander's parking dataset for algorithm performance evaluation.
• Assessed prediction accuracy and information comparability across different algorithms.

Main Results:

• Less complex algorithms (Decision Tree, Random Forest, KNN) outperformed complex ones (Multilayer Perceptron) in prediction accuracy.
• The observed accuracy advantage for simpler models held true regardless of dataset size.
• Provided a method for Top-K parking space recommendations based on vehicle proximity.

Conclusions:

• Simpler ML/DL algorithms are highly effective for parking availability prediction.
• The choice of algorithm significantly impacts prediction accuracy in this domain.
• The study contributes practical insights for intelligent transportation systems and smart city initiatives.