Exploring Interpretable LSTM Neural Networks over Multi-Variable Data

TL;DR

This paper introduces an interpretable LSTM framework for multi-variable time series prediction, capturing variable-specific dynamics and providing insights into variable contributions, with demonstrated improved accuracy and interpretability.

Contribution

It proposes a novel variable-wise hidden state structure and a mixture attention mechanism for interpretable multi-variable time series modeling.

Findings

Enhanced prediction accuracy on real datasets.

Effective interpretation of variable contributions.

Potential as an end-to-end forecasting and knowledge extraction framework.

Abstract

For recurrent neural networks trained on time series with target and exogenous variables, in addition to accurate prediction, it is also desired to provide interpretable insights into the data. In this paper, we explore the structure of LSTM recurrent neural networks to learn variable-wise hidden states, with the aim to capture different dynamics in multi-variable time series and distinguish the contribution of variables to the prediction. With these variable-wise hidden states, a mixture attention mechanism is proposed to model the generative process of the target. Then we develop associated training methods to jointly learn network parameters, variable and temporal importance w.r.t the prediction of the target variable. Extensive experiments on real datasets demonstrate enhanced prediction performance by capturing the dynamics of different variables. Meanwhile, we evaluate the interpretation results both qualitatively and quantitatively. It exhibits the prospect as an end-to-end framework for both forecasting and knowledge extraction over multi-variable data.