A Practical and Cost-Effective Combination of GPS Data and Machine Learning Tools for Detecting Transportation Modes

This work proposes a novel methodology to determine the modes of transportation used in the city of Cuenca, Ecuador, based on geolocation data and machine learning. For this purpose, 354,096 mobility samples from 40 people are collected via their mobile phones, with the respective identification of the transportation mode used: pedestrian, bicycle, bus, tram, taxi, and private vehicle. These samples are used to train and validate supervised learning architectures: classification trees, weighted k-nearest neighbor classifier, support vector machines, and two-layer neural networks. The classification tree achieved an accuracy of 99.5%, followed by 95.7% for the KNN model, 94.0% for SVM, and 93.2% for the BNN model. The trained, validated, and tested classification tree was applied to 110,242 samples obtained from the random mobility of 40 people, generating optimistic results. The findings indicate that the most used mode of transportation is the bus, followed by taxis and private vehicles. Pedestrians and bicycles, as well as trams, are predominantly used in the city center, while private transportation is more commonly used in rural areas. The obtained model can determine mobility patterns in the city, allowing for the effective establishment of origin-destination matrices. This facilitates public transportation planning, promotes alternatives to traditional mobility, and addresses the current problem of increasing traffic congestion, allowing for the reduction of pollutant emissions and mobilization costs, and aiding in the design of a new mobility plan.