[Paper Review] Short Term Prediction of Parking Area states Using Real Time Data and Machine Learning Techniques

Jesper Provoost, Luc Johannes Josephus Wismans+3 more|University of Twente Research Information|Nov 29, 2019

Smart Parking Systems Research20 references4 citations

TL;DR

This paper proposes a real-time machine learning model for short-term prediction of parking area occupancy, in- and outflux up to 60 minutes ahead using historic and live data from Arnhem, Netherlands. A neural network outperforms random forests and a naive seasonal model, achieving over 150% performance gain at 60-minute horizon, with real-time occupancy being the most influential predictor.

ABSTRACT

Public road authorities and private mobility service providers need information derived from the current and predicted traffic states to act upon the daily urban system and its spatial and temporal dynamics. In this research, a real-time parking area state (occupancy, in- and outflux) prediction model (up to 60 minutes ahead) has been developed using publicly available historic and real time data sources. Based on a case study in a real-life scenario in the city of Arnhem, a Neural Network-based approach outperforms a Random Forest-based one on all assessed performance measures, although the differences are small. Both are outperforming a naive seasonal random walk model. Although the performance degrades with increasing prediction horizon, the model shows a performance gain of over 150% at a prediction horizon of 60 minutes compared with the naive model. Furthermore, it is shown that predicting the in- and outflux is a far more difficult task (i.e. performance gains of 30%) which needs more training data, not based exclusively on occupancy rate. However, the performance of predicting in- and outflux is less sensitive to the prediction horizon. In addition, it is shown that real-time information of current occupancy rate is the independent variable with the highest contribution to the performance, although time, traffic flow and weather variables also deliver a significant contribution. During real-time deployment, the model performs three times better than the naive model on average. As a result, it can provide valuable information for proactive traffic management as well as mobility service providers.

Motivation & Objective

• To develop a real-time predictive model for urban parking area states using available data sources.
• To evaluate the performance of machine learning models in forecasting parking occupancy, in- and outflux up to 60 minutes ahead.
• To identify the most influential input variables for accurate prediction.
• To compare the predictive performance of neural networks, random forests, and a naive seasonal random walk model.
• To assess the sensitivity of prediction accuracy to increasing prediction horizons.

Proposed method

• The model uses publicly available historic and real-time data, including occupancy rates, traffic flow, time of day, and weather conditions.
• A feedforward neural network and a random forest model are trained to predict parking state variables at multiple time horizons (up to 60 minutes).
• Model training and evaluation are based on a real-world case study in Arnhem, Netherlands, using a temporal cross-validation approach.
• Feature importance is assessed using permutation-based methods to determine the contribution of each input variable.
• Performance is measured using standard metrics such as RMSE and R² across different prediction horizons.
• The naive seasonal random walk model serves as a baseline for comparison.

Experimental results

Research questions

• RQ1 How accurately can machine learning models predict short-term parking area occupancy and in- and outflux using real-time data?
• RQ2 Which machine learning model—neural network or random forest—performs better for parking state prediction across varying prediction horizons?
• RQ3 What is the relative contribution of real-time occupancy, time, traffic flow, and weather to prediction performance?
• RQ4 How does prediction accuracy degrade as the forecast horizon increases from 5 to 60 minutes?
• RQ5 To what extent does predicting in- and outflux offer higher performance gains compared to occupancy prediction alone?

Key findings

• The neural network model outperforms the random forest and naive seasonal random walk models on all performance metrics, with a 150% performance gain at a 60-minute prediction horizon.
• Predicting in- and outflux is significantly more challenging than occupancy prediction, yielding a 30% performance gain over the baseline, though it is less sensitive to prediction horizon.
• Real-time occupancy rate is the most influential predictor, contributing more than time, traffic flow, or weather variables.
• The model performs three times better than the naive model on average during real-time deployment.
• Performance degrades with increasing prediction horizon, but the neural network maintains strong predictive power even at 60 minutes.
• The inclusion of non-occupancy variables like traffic flow and weather improves model performance, but occupancy remains the dominant factor.