What went wrong and when?
Problem being addressed
Multivariate time series models are poised to be used for decision support in high-stakes applications, such as healthcare. In these contexts, it is important to know which features at which times most influenced a prediction.
A general approach for assigning importance to observations in a multivariate time series. By explicitly modeling the temporal behavior of the signals, the model is able to determine the relative feature importance over time and find the most influential time point for every feature. The importance of each observation is evaluated by its counter-factual effect on the predictive distribution. This method provides explanation at observation level (every feature at every time step) efficiently.
Advantages of this solution
The suggested approach generates the most precise explanations. The method is inexpensive, model agnostic, and can be used with arbitrarily complex time series models and predictors.
Solution originally applied in these industries
Possible New Application of the Work
Time series models are often used to make financial forecasts; however, these forecasts are often inaccurate due to the nature of the data. Applying the algorithms that can also differentiate the importance of each observation may significantly improve this accuracy.
Sales predictions are based on time series and marketeers can definitely benefit from a cheap and adjustable model that allows them to have a better understand the decision making process and come up with better predictions.
Source URL: #############