Forecasting trade balance dynamics: a machine learning approach. Case study of Zimbabwe

Author

Manjengwa, Method

Title

Forecasting trade balance dynamics: a machine learning approach. Case study of Zimbabwe

Abstract

This study employed a quantitative inquiry of the plan that sought to model Zimbabwe's exchange urgency using a 15-year time-series dataset from January 2010 to December 2024. We obtained monthly exchange urgency data from national statistics agencies as they have a comprehensive check outline to capture economic cycles and seasonal variations. Our study examined the forecasting accuracy of classical and machine learning techniques such as (SARIMA), Long Short-Term Memory (LSTM) networks, Feedforward Neural Networks (FFNN), Back Vector Machines (SVM), and Extreme Gradient Boosting (XGBoost). We properly planned the data set by making it stationary using normalisation and differencing. The models were trained and tested using time-based partitioning. Our preferred evaluation tools were Root Mean Squared Error (RMSE) and Mean Absolute Error (MAE) as this allows us to compare the forecasting performance of each show. We concluded that deep learning models, and many particularly LSTM, may lessen the required models on economic forecasting projects where cyclical and seasonal factors must be considered. We recommend that economic planners and lawmakers adopt LST models to improve accuracy of exchange of plans and macroeconomic modelling. We further propose that future models combine hybrid and actual cases.

Date

June 2025

Publisher

BUSE

Keywords

Zimbabwe's Exchange Urgency

Time-Series Analysis

Seasonal Variations

Macroeconomic Modeling

Exchange Rate Forecasting

Supervisor

Dr. T. W. Mapuwei