Magnitude and Rotation Invariant Detection of Transportation Modes with Missing Data Modalities

TL;DR

This paper introduces a robust, rotation- and magnitude-invariant machine learning approach for transportation mode detection using incomplete phone sensor data, addressing distribution shifts between training and validation sets.

Contribution

The work presents a novel rotation-invariant aggregation method and emphasizes the importance of z-normalization for spectral features in transportation mode detection.

Findings

Rotation-invariant aggregation outperforms rotation-aware features.

Z-normalization enhances spectral feature robustness.

Using only magnitude vectors yields poor performance.

Abstract

This work presents the solution of the Signal Sleuths team for the 2024 SHL recognition challenge. The challenge involves detecting transportation modes using shuffled, non-overlapping 5-second windows of phone movement data, with exactly one of the three available modalities (accelerometer, gyroscope, magnetometer) randomly missing. Data analysis indicated a significant distribution shift between train and validation data, necessitating a magnitude and rotation-invariant approach. We utilize traditional machine learning, focusing on robust processing, feature extraction, and rotation-invariant aggregation. An ablation study showed that relying solely on the frequently used signal magnitude vector results in the poorest performance. Conversely, our proposed rotation-invariant aggregation demonstrated substantial improvement over using rotation-aware features, while also reducing the feature vector length. Moreover, z-normalization proved crucial for creating robust spectral features.