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

Jachero, Bryan; Bermeo, Karina · 2025 · Crossref

DOI: 10.1007/978-3-031-87065-1_18

archive: archived pipeline: cataloged verified

Get this paper ↗ (DOI — opens at the source; we link to it, we don't host it)

Summary

This study addresses the challenge of accurately identifying transportation modes in Cuenca, Ecuador, to support sustainable urban mobility planning. Traditional methods for collecting origin-destination data, such as roadside interviews and daily travel surveys, are costly, time-consuming, and prone to bias. To overcome these limitations, the authors propose a cost-effective methodology combining GPS data from mobile devices with machine learning algorithms. The research aims to classify six specific transportation modes—pedestrian, bicycle, tram, bus, taxi, and private vehicle—to generate reliable mobility patterns that can inform public transport planning and reduce traffic congestion. The methodology involved collecting 354,096 mobility samples from 40 participants using Android mobile devices equipped with a background application. Data was recorded at one-second intervals, capturing latitude, longitude, altitude, and speed. From this raw data, the researchers derived key predictors including longitudinal acceleration, jerk, aerodynamic resistance, slope resistance, rolling resistance, inertia resistance, total resistance, traveled distance, and stop times. Four supervised machine learning models were trained and validated using these features: a weighted k-nearest neighbor (KNN) classifier, a support vector machine (SVM), a two-layer neural network (BNN), and a classification tree (DT). The dataset was split into 70% for training, 15% for validation, and 15% for testing. The results demonstrated that the classification tree model achieved the highest performance, with an overall accuracy of 99.5%. This significantly outperformed the other models, which achieved accuracies of 95.7% for KNN, 94.0% for SVM, and 93.2% for the BNN. The classification tree showed particularly high precision for bicycle, bus, and private vehicle modes, with minimal difficulty in distinguishing classes except for pedestrians and trams. When applied to a separate random dataset of 110,242 samples, the model revealed distinct spatial usage patterns: buses were the most utilized mode overall (39.35% of samples), followed by taxis (28.82%) and private vehicles (19.52%). Pedestrian, bicycle, and tram usage was concentrated in the city center, while private vehicles and taxis dominated rural areas. The significance of this work lies in its ability to provide a high-accuracy, low-cost alternative to traditional mobility surveys. The validated model can effectively establish origin-destination matrices, facilitating better public transportation planning and the design of new mobility plans. By enabling precise detection of transportation modes, the approach supports efforts to reduce pollutant emissions and mobilization costs. However, the authors note that the current sample size does not represent the entire city population, recommending future replication with larger datasets to enhance generalizability.

Provenance

The full processing record for this entry. Every stage of this paper's journey through the pipeline is logged — what ran, with which tool and model, how many attempts it took, and when it last completed.

Summary generated by qwen3.6-27b-prismaquant on 2026-06-26; verification: verified.

Topics

Ranked by relevance to this paper. Hover a topic for its definition.

• telematics crash prediction