Support vector machines and k-means to build implementation areas of bundling hubs

Abstract

City transportation has three basic components that create the essential environment for its functioning and the social welfare namely infrastructure, operational assets, and management policies. The key focus of this article is on understanding long-term distribution of transport demand in order to build bundling networks. To achieve this aim, we provide a hybrid machine-learning approach using a combination of several clustering and forecasting algorithms that are considered efficient given the key performance indicators obtained. This approach involves combining two types of algorithms: clustering and prediction algorithms. Based on simulated benchmarks, results indicated that the clustering phase is still appropriate using the k-means algorithm. To improve the k-means results, we measured 30 validation indices to estimate the number of clusters. In so doing, not only does it want to validate the clusters but also to identify the optimal k. To evaluate forecast accuracy in the demand prediction phase, we used the standard key performance indicators, namely MSE, RMSE, MAPE and R². The SVM algorithm has been judged as the most efficient prediction algorithm based on average values of the obtained metrics.