PREDICTION ACCURACY OF NIGERIAN MILITARY EXPENDITURE: MLR, ARIMAX, AND ANN MODELS IN STATISTICAL AND MACHINE LEARNING FRAMEWORKS

Abstract

Accurately predicting military expenditure is crucial for budgetary planning and national security. However, traditional forecasting methods often struggle to capture the complex dynamics of military spending. This study investigates the potential of statistical methods and machine learning to improve the accuracy of Nigerian military expenditure prediction. We compare the performance of three widely used models: multiple linear regression (MLR), autoregressive integrated moving average with exogenous variables (ARIMAX), and artificial neural networks (ANN). Using historical data on Nigerian military expenditure, GDP, and relevant economic indicators, we train and evaluate each model's prediction accuracy. We also employ statistical tests to assess the normality of residuals in the two distinct models of MLR and ARIMAX. Our findings indicate that the machine learning model, particularly ANN, significantly outperforms MLR in terms of prediction accuracy. ARIMAX shows promising results but lags behind ANN. We attribute the superior performance of the ANN to its ability to capture non-linear relationships and complex patterns in the data. This study adds to the body of knowledge by highlighting how machine learning methods can be used to increase the accuracy of military expenditure forecasts. Furthermore, the specific focus on Nigeria provides valuable insights into the unique dynamics of military spending in a developing country.