Investigating Methods of Determining Number of Hidden Units in Deep Learning for Taxi Recommender System
Undarmaa Chinzorig, Ha Yoon Song, Jun T. Park · 2019
With modern ubiquitous computing environments, recommender systems have become a major part of modern intelligent services. Taxi recommender system allows both passengers and drivers to minimize waiting times and obtain the current location of each other. In this paper, we present a taxi recommender system based on deep learning for catching taxis. In deep learning technique, random selections of hyperparameters lead to overfitting or underfitting problems in prediction or classification. In particular, determining number of hidden units is one of the critical issues facing research. Therefore, we investigated how hidden units affect the performance of deep learning for taxi recommender system and compared the results of these existing methods for determining number of hidden units. Deep Learning algorithms, such as Deep Neural Network (DNN), which have been successfully used, were employed for classifying road segments. Finally, our proposed system can spot regions, where a passenger can catch a taxi within walkable distance.