A passenger detection method is required for smart public transportation systems. Such a method would enable control and orchestration of transit routes and schedules. To validate the method we proposed to differentiate people who appear to get into a transportation vehicle from others, we used an imbalanced WiFi-based dataset. However, the imbalanced dataset which includes majority and minority classes in the training set affects the performance of machine learning algorithms. Unreliable samples in the majority class can perturb the minority class. This paper proposes a method to efficiently classify imbalanced datasets. Then we explore the impact on classification performance of the imbalance observed in the WiAR dataset in relation to feature selection in a proposed machine-learning-based classification algorithm. The results show that the proposed method improves the F1-score performance for the minority class from 46% to 95%.