SMU Data Science Review


Current nonlinear time series methods such as neural networks forecast well. However, they act as a black box and are difficult to interpret, leaving the researchers and the audience with little insight into why the forecasts are the way they are. There is a need for a method that forecasts accurately while also being easy to interpret. This paper aims to develop a method to build an interpretable model for univariate and multivariate nonlinear time series data using wavelets and symbolic regression. The final method relies on multilayer perceptron (MLP) neural networks as a form of dimensionality reduction and the PySR algorithm to determine the symbolic relationships. It also explores use cases for using the discrete wavelet transformation to extract information from the dataset.

Creative Commons License

Creative Commons Attribution-Noncommercial 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial 4.0 License