Bus arrival time prediction using artificial neural network - The case of istanbul

Travel time prediction is essential to maintain service efficiency and quality for today's travel agencies. This paper aims to propose an artificial neural network (ANN) by utilizing real time and historical data together and focusing on link travel time prediction in order to estimate bus arrival time to bus stops. We trained the time prediction ANN using link-based and stop-based variables such as previous link travel time, traffic conditions, etc. Multilayer-Perceptron with back propagation model is chosen as ANN topology due to proven estimation success in the literature. Two months real data for four different bus routes in Istanbul are classified before training ANNs according to time-of-day, day-of-week and link resistance. Suggested tool can provide basic supportive data for bus route planning and timetable design systems. Since all used parameters can be measured online, it can be integrated to GPS data collecting system to provide real time information. Together with real time data, agency can offer more reliable information about bus arrival time on bus stops or on mobile applications which lead quality and customer satisfaction.