# Neural Decomposition of Time-Series Data for Effective Generalization

## TL;DR

This paper introduces Neural Decomposition, a neural network method that combines sinusoidal and nonperiodic units to analyze and extrapolate diverse time-series data, demonstrating superior performance over existing techniques.

## Contribution

The paper presents a novel neural network approach that effectively decomposes and extrapolates time-series data using sinusoidal units with regularization and careful initialization.

## Key findings

- Outperforms LSTM, ARIMA, and other models in forecasting accuracy.
- Effectively generalizes across diverse real-world datasets.
- Uses a simple, well-regularized model for robust time-series analysis.

## Abstract

We present a neural network technique for the analysis and extrapolation of time-series data called Neural Decomposition (ND). Units with a sinusoidal activation function are used to perform a Fourier-like decomposition of training samples into a sum of sinusoids, augmented by units with nonperiodic activation functions to capture linear trends and other nonperiodic components. We show how careful weight initialization can be combined with regularization to form a simple model that generalizes well. Our method generalizes effectively on the Mackey-Glass series, a dataset of unemployment rates as reported by the U.S. Department of Labor Statistics, a time-series of monthly international airline passengers, the monthly ozone concentration in downtown Los Angeles, and an unevenly sampled time-series of oxygen isotope measurements from a cave in north India. We find that ND outperforms popular time-series forecasting techniques including LSTM, echo state networks, ARIMA, SARIMA, SVR with a radial basis function, and Gashler and Ashmore's model.