MODELING AND FORECATING INFLATION RATE IN SRI LANKA USING CLASSICAL TIMESERIES, NEURAL NETWORK AND ARIMA- NEURAL HYBRID MODELS

This study explores the efficacy of various forecasting models in predicting monthly inflation rates in Sri Lanka, utilizing historical data spanning from 1988 to 2023. The research primarily focuses on the application and comparison of traditional time series models, machine learning techniques, and a novel hybrid model combining the strengths of both approaches. The key models analyzed include the Autoregressive Integrated Moving Average (ARIMA), Artificial Neural Networks (ANN), LSTM, PZM, and the proposed Hybrid Model (HB), which integrates ARIMA and ANN to capture both linear and nonlinear patterns in the data. The study begins with a detailed examination of classical time series models, particularly ARIMA, to assess their ability to forecast inflation. However, recognizing the limitations of these models in capturing the nonlinear dynamics often present in economic data, the research extends to machine learning models, including Lasso Regression (LR), Bayesian Ridge Regression (BRR), Support Vector Regression (SVR), and Random Forest Regression (RF). These models are rigorously evaluated using cross- validation techniques such as Walk Forward Validation (WFV) and K-fold Cross-Validation (KCV). To address the identified shortcomings of individual models, the study introduces a hybrid approach (HB Model) that leverages the predictive power of ARIMA for linear components and ANN for nonlinear patterns. The proposed HB Model is systematically worked out and empirically tested showing less MAPE than standalone models establishing the proposed HB Model. The outcome of this study shows that the HB Model has greater forecasting performance than conventional ARIMA methodologies and also surpasses standalone machine learning algorithms with a MAPE of 7.61% Of the test data, thus stressing its stability when faced with actual forecasts. The study concludes that hybrid models, particularly the HB Model, offer a promising solution for improving inflation forecasts in complex economic environments like Sri Lanka. Furthermore, the research underscores the potential of integrating advanced data processing techniques with traditional models to enhance predictive accuracy, providing valuable insights for policymakers and economists in managing inflation and economic planning.