Expectation maximization imputation in forecasting the rate of unemployment in South Africa

Type Thesis or Dissertation - Research Report
Title Expectation maximization imputation in forecasting the rate of unemployment in South Africa
Publication (Day/Month/Year) 2022
URL https://www.riteshajoodha.co.za/sitepad-data/uploads/2023/10/thabetheandile_24207_4431567_2092108_Re​search_Report.pdf
According to statistics from the fourth quarter of 2021, South Africa’s unemployment rate grew from 34.9 percent in the previous period to 35.3 percent. Unemployment rates are widely recognized as crucial indicators of a nation’s labor market success. Particularly during difficult economic times and recessions, it is an economic indicator that is closely watched. Previous studies have used a variety of data imputation techniques to impute missing values in the SARB and BER data sets, including constant imputation, last known value imputation, multivariate imputation using chained equations, k-nearest neighbor imputation, forward imputation, and mean value imputation. Last known value imputation imputation produced the best results. Past research improved the prediction of the unemployment rate in South African through using machine learning approaches over traditional statistical models, deep learning, feature selection and engineering but not through data imputation. This study investigated the performance of expectation maximization imputation against last known value imputation to better forecast the unemployment rate South African unemployment rate. Traditional statistical models performed better when the expectation maximization imputed data was used. The deep learning methods performed better when the last known value data set was used. This study demonstrated that there is a role for expectation maximization imputation in the prediction of the rate of unemployment in South Africa when traditional statistical methods are used.

Related studies