Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

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

Jeroen Van Der Donckt, Jonas Van Der Donckt, Sofie Van Hoecke

TL;DR

This work presents the solution of the Signal Sleuths team for the 2024 SHL recognition challenge, focusing on robust processing, feature extraction, and rotation-invariant aggregation, and z-normalization proved crucial for creating robust spectral features.

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.

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

TL;DR

This work presents the solution of the Signal Sleuths team for the 2024 SHL recognition challenge, focusing on robust processing, feature extraction, and rotation-invariant aggregation, and z-normalization proved crucial for creating robust spectral features.

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.
Paper Structure (14 sections, 4 equations, 4 figures, 2 tables)

This paper contains 14 sections, 4 equations, 4 figures, 2 tables.

Table of Contents

  1. Introduction
  2. SHL 2024 Challenge dataset
  3. Exploratory Data Analysis
  4. Algorithm pipeline
  5. Processing
  6. Feature Extraction
  7. Model
  8. Postprocessing
  9. Rotation Invariant Aggregation
  10. Experimental results and discussion
  11. Impact Distribution Shift
  12. Final Model
  13. Improving the Test Score
  14. Conclusion

Figures (4)

  • Figure 1: Label distribution for train and validation dataset (column 1-2) and number of unique labels per window (column 3-4).
  • Figure 2: Data distribution (of the standard deviation - STD) for the train and validation dataset.
  • Figure 3: Exploratory time series visualization of phone movement data from train set (location = Hips) and labels.
  • Figure 4: Confusion matrix for the final model (i.e., rot_inv$_{stat2}$ + SMV + SMV$_{dt2}$) on validation dataset (including all 4 phone locations). The validation predictions were obtained using the 3-fold MV method, as described in Section \ref{['sec:postprocess']}.