Transport Logistic Application: Train’s Adherence Evolution and Prediction Based on Decision Tree and Markov Models

In this paper it is presented a method for modelling adherence and prediction of train evolution based on classification tree and Markov Models. In day-to-day of railway operational management of traffic it is recurrent to face a deviation from the theoretic planned schedule to a certain space-time. When such deviation happens, what we also call non-adherence, there is a need for a quick decision-making to avoid, on the one hand, that primary initial non-adherence leads to a whole cascade of secondary non-adherence of other trains over the planned schedule, on the other hand that event such as a train reaches another may occurs, what is a considerable issue. We present an effective mixed approach compounded of a stochastic process (Markov Models) and classification tree for estimating and predicting the reliability of the adherence of planned and realized traffic schedule of not only rail freight transport, but also public transport, considering some parameters that may cause deviation such as weather, environmental conditions, and so on. The experiments on real data for trains of Brazilian regions show that our model is fairly realistic and deliver good results in short processing time. Moreover, the model may run for an on-line scenario where updated data are massively collected from monitoring sensors. Withal, the prediction’s accuracy along with the evolution of probability distributions regarding all events over time have been evaluated. Therefrom the approach reveals to be good for predicting for train traffic evolution based on historic data and monitoring collected data as well as weather and environmental ones. As a result we obtain the increasing reliability of about 74% of prediction.