Comparative Study of Machine Learning (ML) and Conventional Time Series Methodologies in Modelling the Exports Trade of Pakistan

Abstract

Export trade is a pivotal driver of economic growth and stability in any nation, including Pakistan. Accurate modelling of export trade holds immense significance as it allows for informed decision-making and strategic planning. By employing advanced techniques like machine learning alongside traditional time series models, we can gain deeper insights into the dynamics of exports, anticipate trends, and adapt policies and strategies accordingly. This study focuses on comparative study of different machine learning and time series models in forecasting the exports trade of Pakistan. The dataset used in this study spans from 1972 to 2021, containing yearly data on exports (% of GDP) for Pakistan. This study focuses on employing various machine learning models, including the MLP model, ELM model, and classical time series models like ARIMA and Exponential smoothing, to model and forecast the exports of Pakistan. The model that best met the KPI criteria was chosen as the optimal candidate for predicting the behaviour of the exports trade data. The outcomes of the study concluded that MLP model outperformed in performance metrics, with MSE = 0.49, RMSE = of 0.70, MAE = 0.53, and a MAPE = 4.27. among all alternative models, including the Extreme Learning Machine model, ARIMA model, and Simple Exponential Smoothing. Our study highlights the effectiveness of the Multi-layer Perceptron (MLP) model as a valuable tool for forecasting exports. Its accuracy empowers policymakers to make well-informed decisions, ensuring a more sustainable and prosperous economic future.