The credibility evaluation of the trajectory clustering results using a user- defined similarity

Evaluation of the cluster analysis results in spatio-temporal trajectories is more sophisticated than the same procedure in other data. Measuring clusters’ compactness and separateness requires defining an appropriate similarity function. Similarity definitions in trajectories are diverse and application-based. Positional similarity and the similarities of speed and direction, as elemental features of moving objects, are fundamental concepts in the trajectory similarity definition. In this paper, we present a new framework for evaluating trajectory clustering results based on the expert’s opinion on the definition of similarity. Specifically, the meaning of similarity is defined by the experts using the AHP method and based on the application context. Moreover, we propose a new index, which is utilized in estimating the optimal cluster number. Based on the obtained results, taking the application and the data structure into consideration is very influential in the evaluation process. To verify that they are not random, the one-way ANOVA test is carried out at the confidence interval of 95% to provide the significance test of the results.