A Deep Learning Approach to Predict Parking Occupancy using Cluster Augmented Learning Method

This paper proposes a deep learning model for block-level parking occupancy prediction. The proposed model leverages Convolutional Neural Nets (CNN) to extract spatial relations of traffic flow and stacked LSTM autoencoder to capture temporal correlations. The paper also introduces Clustering Augmented Learning Method (CALM) which is based on concept of simultaneous heterogeneous clustering and regression learning to learn deep feature representations of spatio-temporal data obtained using the proposed embedding. The regression model that is considered in this work has a Feedforward Neural Net(FNN) architecture. With the aim of improving the accuracy of the regression model, CALM iterates between clustering and learning to form robust clusters, thereby leveraging the learning process of the FNN. A dissimilarity measure is proposed based on the weights of each feature of input data in the regression model to form the clusters. During each iteration of learning and clustering, a classifier is used to predict the current cluster labels, and the cluster belonging probabilities are used to control the subsequent re-estimation of cluster centers. Computational experiments on San Francisco Parking data set reveals that proposed approach CALM outperforms other baseline methods including multi-layer LSTM and Lasso with testing MAPE of 7.8% when predicting block-level parking occupancies. The findings highlight that the way in which the spatio - temporal information is exploited in order to aggregate the analogous data into clusters makes a significant difference and demonstrate the effectiveness of the model in processing parking data.