Using Feature Engineering Techniques to Improve Multivariate Model Forecasting of the South African Unemployment Rate

Abstract

The unemployment rate in South Africa has been increasing yearly. Between January and March 2021, the unem-ployment rate has risen to 32.6 &#x0025;. This means that there are more than 7.8 million South Africans who are officially unemployed [1]. This paper explores feature engineering techniques that can be applied to improve the forecasting ability of multivariate machine learning forecasting models. Research has shown the ability of modern machine learning methods to forecast the unemployment rate in South Africa accurately. Few kinds of research have been conducted on improving these models. Feature engineering is one of the many ways that can be applied to improve the performance of these models. In this paper, we are looking at several feature engineering techniques such as combining, performing statistical calculations, and transforming features. Using SVR as the base model, the feature engineering techniques managed to improve the <tex>$R^{2}$</tex> by over 80 &#x0025;. This paper presents feature engineering as a viable solution to improving the forecasting ability of multivariate forecasting machine learning models.