Title: A low dimensional model for bike sharing demand forecasting

Abstract: Big, transport-related datasets are nowadays publicly available, which makes data-driven mobility analysis possible. Trips with their origins, destinations and travel times are collected in publicly available big databases, which allows for a deeper and richer understanding of mobility patterns. This paper proposes a low dimensional approach to combine these data sources with weather data in order to forecast the daily demand for Bike Sharing Systems (BSS). The core of this approach lies in the proposed clustering technique, which reduces the dimension of the problem and, differently from other machine learning techniques, requires limited assumptions on the model or its parameters. The proposed clustering technique synthesizes mobility data quantitatively (number of trips) and spatially (mean trip origin and destination). This allows identifying recursive mobility patterns that - when combined with weather data - provide accurate predictions of the demand. The method is tested with real-world data from New York City. We synthesize more than four million trips into vectors of movement, which are then combined with weather data to forecast the daily demand at a city-level. Results show that, already with a one-parameters model, the proposed approach provides accurate predictions.