A large international Car-Park manager, which operates more than 5.000 parking garages/facilities would like to be able to predict, when their facilities will be occupied, to which extend (percentage). The data the company has available today is historical data on occupancy by the hour of the day (as well as weather data).
Nordic Analysis delivered a concept for a scalable solution and demonstrated how their Artificial Intelligence (AI) solution – Cayce Alpha– can predict occupancy rate by the hour. The project was based on the example of a car park facility in Germany.
For predicting the occupancy, Nordic Analysis build a model for analysing a sequence of different observations that occur sequentially in time (time series). Among other, the sequence included occupancies of car parking, discrete measurements retrieved from weather stations, social and economic indicators. The goal was to find out different dependencies between observations that are close to each other, and to formulate the dependencies in a form of a mathematical model.
We further performed a sensitivity analysis, which gave us information on which input factors had a significant influence on the car-park occupancy. The report shows that external events, historical data set along with traffic information and data from Twitter are having the highest influence on the prediction of the occupancy rate of the selected car park.
Comparing to statistical and other machine learning methods, the solution developed by Nordic Analysis allows easily to synthesize regression models of any level of complexity. Combining variety of numerical and textual data sources, Cayce Alpha performs validation of the data diversity leveraging its built-in Sensitivity Analysis service.
Transfer learning-based functionality is another key feature of Cayce Alpha. It simultaneously adapts and optimizes model architecture whilst learning underlying data sets. In the project, Nordic Analysis tested several machine-learning based approaches which are well known in area of time series forecasting: linear regression, regression trees, support vector machines (SVM) and neural networks. The best results were achieved with an RNN (recurrent neural network) based on Cayce Alpha, the AI technology from Nordic Analysis.