Data Fusion for Deep Learning on Transport Mode Detection: A Case Study

TL;DR

This paper evaluates various data fusion methods for transport mode detection using real-world data, revealing that simple late fusion often outperforms more complex approaches and emphasizing the importance of sensor and data representation choices.

Contribution

The study provides a comprehensive experimental comparison of data fusion techniques in transport mode detection, highlighting the effectiveness of late fusion and the impact of sensor and data representation choices.

Findings

Late fusion outperforms complex methods.

2D spectrograms with logarithmic frequency axis improve accuracy.

Sensor choice significantly affects detection performance.

Abstract

In Transport Mode Detection, a great diversity of methodologies exist according to the choice made on sensors, preprocessing, model used, etc. In this domain, the comparisons between each option are not always complete. Experiments on a public, real-life dataset are led here to evaluate carefully each of the choices that were made, with a specific emphasis on data fusion methods. Our most surprising finding is that none of the methods we implemented from the literature is better than a simple late fusion. Two important decisions are the choice of a sensor and the choice of a representation for the data: we found that using 2D convolutions on spectrograms with a logarithmic axis for the frequencies was better than 1-dimensional temporal representations.