Comprehensive Predictions of Tourists' Next Visit Location Based on Call Detail Records Using Machine Learning and Deep Learning Methods.

Recent developments in data mining and machine learning have helped to solve many issues in prediction and recommendation. In this project, we run a comprehensive study on individual behavior patterns from call detail records (CDR) data to predict touristsu0027 future stops. Multiple classification algorithms are employed, including Decision Tree, Random Forest, Neural Network, Naive Bayes and SVM. In addition, a Recurrent Neural Network-Long Short Term Memory (LSTM) that is ordinarily applied to language inference problems is tested. Surprisingly, we find that LSTM provides us with the best prediction (94.8%), while Random Forest/Neural Network give the second best (85%). Our investigation suggests that the memory-dependence property of LSTM architecture gives it great expressive power to model our time-series location data, making it an outstanding classifier.