The inherent feature of artificial neural networks is an efficient information processing system. It has been successfully applied to various transportation problems of classification, prediction and optimization. The network structure is composed of input layer, output layer and hidden layer. Owing to lack of specific criteria, some input layer variables may be less relevant to the desired output. This would increase the difficulty of data collection and network operations. This study investigates the relationships between input and output elements using the contribution graph approach. Transit containers forecast in Kaohsiung port is employed for illustration. Significant inputs relationships are identified easily from the network. Based on the predicted volume and sensitivity analysis, the proposed approach is confirmed an efficient way to utilize the neural networks.