Multivariate Time Series Forecasting using XGBoost
Introduction
Univariate Time Series analysis is fine for simpler time dependent variables however, various enterprise data are dependent on multiple data drivers varying over time. These data drivers can be internal to organization or can be external as well. To forecast these enterprise data often we use Multivariate Time Series analysis.
Prerequisites
Before we start, please go through my previous blogs:
- 10 Time Series Forecasting Methods We Should Know
- Time Series Forecasting — Parallel Processing using Pandas Function APIs
Approach
There are multiple multivariate forecasting methods available like — Pmdarima, VAR, XGBoost etc. In this blog, we’ll focus on the XGBoost (Extreme Gradient Boosting) regression method only.
First we’ll use AR (AutoRegressive) model to forecast individual independent external drivers. Once these univariate time series forecasts are available we’ll apply the scikit-learn API for XGBoost regression to forecast the dependent variable.
