Predicting user’s next location using machine learning algorithms

The reliability on smartphones has been increased in a tremendous way in the last decade, and this incensement brings us nothing but data to investigate. Furthermore, the ability to predict user’s next location plays a vital role in Location-Based Services and recommender systems. This paper is seeking to predict the user’s next location based on their spatial background using machine learning methods like Artificial Neural Networks and Classification methods like K-Nearest Neighbors (KNN), Support Vector Machine and Decision Tree. The suitable method is then chosen through their comparison. The data used in this research is from active users provided from the Geolife dataset from the city of Beijing. The best result is obtained from the Weighted K-Nearest Neighbors (KNN) method, with an overall accuracy of 91.98 percent. Additionally the dependency between observed data and model prediction is determined. A concept called Routineness is also introduced, which shows the predictability and anomalies of each users’ behavior, based on the difference in prediction methods from their spatial and temporal background. A comparison is also done with similar researches that used the same data to evaluate the sufficiency of the methods. The computation shows the proposed method predicts behavior 2.72% more accurate than similar ones.