Predicting South African GDP growth using machine learning approaches

This study uses secondary data from Stats SA (https://www.statssa.gov.za/…) to predict South Africa’s real GDP using ML. The dataset is structured on an annual frequency basis for training ML models. In this dataset, there are 20 years data of various economic variables. There are 11variables which are agriculture, forestry and fishing, Mining and quarrying, Manufacturing, Electricity, gas and water, Construction, Trade, catering and accommodation, Transport, storage and Communication, Finance, real estate and business services, General government services, Personal services and GDP at market prices. Since the goal is to predict GDP growth rates, the GDP at market prices is the dependent feature and other are independent features. Data was imported from a structured file that included historical economic variables data. Rows with continuous missing data from 1993 to 2023 were deleted to ensure temporal integrity, as imputation was regarded unreliable for extended gaps. This extended period of missing data made those rows incomplete and unreliable for time series analysis, which requires continuity and temporal integrity. Maintaining a thorough chronological record is critical in time series studies for accurately identifying trends, seasonality, and other temporal patterns. Retaining such partial records could have added bias in the integrity of the models. To concentrate the analysis on important economic variables, unnecessary columns that added nothing to the core analysis were removed. Thereafter, the variables related to GDP at Market Prices were filtered using data expressed in constant prices. The columns were renamed for easy readability. The dataset was transposed, turning columns into rows, to make time series modelling easier and provide a more logical structure for sequential analysis.