Biography
Biography pending.
Conference contributions 1
FORECASTING NIGERIA'S HEADLINE INFLATION:
A comparative evaluation of SARIMA, LSTM, XGBoost and MLP models
Reliable inflation forecasting is central to monetary policy formulation and macroeconomic management, particularly in structurally volatile economies. This study comparatively evaluates four forecasting models - Seasonal Autoregressive Integrated Moving Average (SARIMA), Long Short-Term Memory (LSTM) networks, Extreme Gradient Boosting (XGBoost) and Multilayer Perceptron (MLP) - applied to monthly headline inflation rate of Nigeria from January 2009 to November 2024 (191 observations). An 80:20 temporal train-test split was applied uniformly across all models. Stationarity was established after second-order differencing and seasonal analysis confirmed an annual cycle at lag 12. MLP achieved the best forecasting performance across all evaluation metrics (RMSE = 1.02, MAE = 0.86, MAPE = 3.62%, R² = 0.975), outperforming other models. XGBoost performed poorly on the test set (RMSE = 8.94, R² = -0.892) despite strong cross-validation performance (CV RMSE = 1.20), a divergence attributed to the structural inflation break of 2022–2025 that exceeded the model's training distribution range. These findings establish MLP as the most accurate standalone model for Nigeria's inflation series and demonstrate that architectural simplicity, when appropriately calibrated, can outperform more complex deep learning architectures on small macroeconomic datasets.