Customer Prediction using Parking Logs with Recurrent Neural Networks

Neural Networks have been performing state of the art for almost a decade now; when it comes to classification and prediction domains. Within last few years, neural networks have been improved tremendously and their performance is even better than humans in some domains, e.g. AlphaGo vs Lee Sedol and Image Net Challenge-2009. It’s a beneficial factor for any parking lot to know that what would be a parking position at any given point in time. If we are able to know in advance that are we going to get parking tomorrow afternoon in a busy super store parking lot, its very beneficial to plan accordingly. In this paper, we predict customer influx in a specific departmental store by analyzing the data of its parking lot. We use this parking data to predict the customer influx and outflux for that parking lot as this parking influx is directly proportional to the customer influx in the store. We use Recurrent Neural Network on the top of two years of historical data. We generate promising results using this dataset by predicting the traffic flow for each hour for next 7 days. We further improve our performance on this dataset by incorporating three more environmental factors along with the parking logs.