Travel time pattern analysis and prediction are essential for achieving better logistics performance in the supply chain. Solid theoretical assumptions based on reliable historical information must be established to analyze travel time; however, access to such information in emerging markets is challenging. Neural networks can learn historical data patterns and are proposed in this study as an artificial intelligence tool to calculate and forecast travel times to develop reliability measurements. Thus, the following measures and indices were used: the percentiles of travel time and the mean, amplitude, skew, buffer, and indices of fluidity and planning time regarding the mean. The obtained data were compared, and a small variation was found between the control and prediction sets. Furthermore, the model did not generate large prediction errors based on the root-mean-square error (RMSE) values. According to the mean difference test results, the hypothesis that the real and forecasted datasets have the same mean was not rejected. Overall, the possibility of predicting travel times using neural networks allows modeling the transportation segments where information is unavailable, thus, preserving travel data anonymity. Finally, helpful conclusions and a proposal for future research are presented.
