A large volume of work has been done to summarise the pattern of a pedestrian’s trajectory and to detect the intention of the pedestrian, based on their previously recorded trajectory. However, little effort has been devoted to implementing the real-time trajectory pattern prediction. This study is an innovative experiment which attempts to develop a system, employing machine learning algorithms to categorise the transportation mode and activity on the fly given the geospatial information from a pedestrian’s smartphone. Firstly, a Gaussian Mixture Model, aimed at predicting the transportation mode as well as a Naive Bayes Model and a Decision Tree Model, both of which aiming to predict a pedestrian’s current activity, are built, using two sets of training data. Then raw data collected from the smartphone is pre-processed and transformed into an ARFF format file. Finally, every record in the ARFF format file, which represents the instantaneous status of the pedestrian, is imported into the model and the result tables, which depict the probability of every transportation mode and activity of the pedestrian’s current status, are shown in the system.