A machine learning-based tourist path prediction

Intelligent recommendations about where to go will be very helpful for personal and commercial travel recommendations. We deal with this kind of recommendations as a prediction problem based on the tourist's historical visiting sequences and supervised machine learning algorithms, namely Random Forests and LambdaMART. We propose a feature set with 56 dimensions from the tourist's historical traveling data. By utilizing the entropy from information theory, all features are ranked. Evaluation results show that when selecting the most important subset of 20 features as our final input of Random Forests, we can get a 4% higher accuracy and a 70% reduction of computation complexity with regard to using the full set of features. A comparison of five different machine learning algorithms, namely Random Forests, LambdaMART, Ranking SVM, ListNet and RankBoost, has been taken on this feature set, results demonstrate that the Random Forests outperforms the other algorithms.