A Deep Learning Based Framework for Local Inflation Direction Prediction to Support Poverty Reduction and Economic Stability
Year: 2026
Type: article
Abstract: Inflation volatility causes significant obstacles to economic stability and poverty reduction, particularly in developing regions. Therefore, accurate local-level inflation forecasting is essential for evidence-based policy design aligned with the Sustainable Development Goals. This study is based on an empirical analysis of monthly inflation data from India over the period 2014 to 2025. Macroeconomic inflation datasets usually have limited data points and strong dependence on past values, which makes model training and generalization difficult. While deep learning models are powerful and can learn complex time-based patterns, using them for inflation prediction is challenging. This study suggests a hybrid directional inflation prediction framework that combines traditional econometric preprocessing with deep learning models that learn patterns over time. These deep learning models are compared with a simpler logistic regression model using experiments that respect the time order of the data. Using historical Consumer Price Index (CPI) and macroeconomic indicators, baseline models such as logistic regression are compared with Long Short Term Memory (LSTM) and Gated Recurrent Unit (GRU) models. Experimental results reveal that the logistic regression model achieves higher accuracy in predicting short-term inflation directions under limited data, indicating its effectiveness as a classification-oriented policy signal. The GRU model achieves comparable accuracy with improved training stability, whereas the LSTM suffers from overfitting. Food and oil price indices significantly influence inflation dynamics. Deep learning models complement rather than replace classical methods. Overall, the framework supports reliable and policy-relevant directional inflation signals. The proposed method can help policymakers make better decisions that support poverty reduction and economic growth (SDG-1 and SDG-8).
Authors: Dey, Arundhuti, Basu, Samrat, Dhar, Suparna
Language: English
Field: Decision Sciences
Domain: Social Sciences
SDGs: Climate action, No poverty
Open Access status: green
OpenAlex ID: W7204440271