The Local Pickup and Delivery Problem (LPDP) has drawn much attention during recent years. In the literature, optimization models and algorithms have been developed to address this problem. However, for some real world applications, the large-scale and dynamic nature of the problem causes some difficulties in getting good solutions within an acceptable time through standard optimization approaches. On the other hand, actual dispatching solutions made by field experts in transportation companies contain embedded useful dispatching rules. This paper presents a general Data Mining-Based Decision System (DMBDS) framework to mimic current dispatch processes and generate solutions for LPDPs by learning from historical data. An application in the intermodal freight industry is presented, where the DMBDS provides good solutions which are comparable to the real solutions provided by expert dispatchers, with respect to a set of key performance indicators.