Many-to-Many Prediction for Effective Modeling of Frequent Label Transitions in Time Series

Description

Code from the paper titled "Many-to-Many Prediction for Effective Modeling of Frequent Label Transitions in Time Series" published at PETRA 2024 conference.

Abstract.

Time series classification is vital in areas like financial forecasting, health monitoring, and human activity recognition. Traditional time-series models segment data into win- dows and assign one label per window, often missing label transitions within those win- dows. This paper presents a many-to-many time-series model using hybrid recurrent neu- ral networks with attention mechanisms. Un- like typical many-to-many models, our ap- proach doesn’t require a decoder. Instead, it employs an RNN generating a label for every input time step. During inference, a weighted voting scheme consolidates overlap- ping predictions into one label per time step. Experiments show our model remains effec- tive on time series with sparse label shifts, but particularly excels in detecting frequent transitions. This model is ideal for tasks de- manding accurate pinpointing of rapid label changes in time-series data.

Setup

Requires

• Python 3.6 or greater
• TensorFlow 2.4 or greater
• Compatible versions of
  • numpy
  • pandas
  • xgboost
  • sklearn
  • matplotlib

Usage

Usage: main.py [-vg] (assuming python3 in /usr/bin/)

v: verbose mode (optional) g: graphing mode (optional)

Configuration settings in src/cfg.py

• run training first with SEC2SEQ=False to get A/P many-to-one for analysis and comparison to many-to-many
• then run this file with cfg.SEC2SEQ=True to get A/P many-to-many for analysis and comparison
• for many-to-one to many-to-many analysis and comparison
  • run hard vote
  • run soft vote without attention
  • run soft vote with attention
  • run stacking
  • compare all five

Citation

@inproceedings{katrompas2024many,
  title={Many-to-Many Prediction for Effective Modeling of Frequent Label Transitions in Time Series},
  author={Katrompas, Alexander and Metsis, Vangelis},
  booktitle={Proceedings of the 17th International Conference on PErvasive Technologies Related to Assistive Environments},
  pages={265--272},
  year={2024}
}