Machine learning-enhanced ARIMA-LSTM model for prediction of monthly district prices: a case study of 22 districts of Tehran Municipality

This study proposes a machine learning-enhanced ARIMA-LSTM framework for predicting monthly housing prices across the twenty-two districts of Tehran, Iran. The model combines ARIMA-based trend stabilisation with LSTM-based nonlinear learning and uses a data-driven machine-learning integration stage to improve city-wide price prediction. To extend the city-wide forecasts to local markets, a district ratio modelling mechanism is used to reconstruct district-level prices from the common Tehran market trend and district-specific relative patterns. The framework also incorporates timely financial proxies, including exchange rates, gold prices, and stock-market indicators, to account for macro-financial volatility in a sanction-exposed and inflationary housing market. The results show that the hybrid ARIMA-LSTM framework provides a comparatively stable overall profile and substantially outperforms the independent models over the longer eight-month horizon, although ARIMA remains slightly more accurate over the shorter horizons. The ratio-based reconstruction provides smoother district-level relative series while preserving localised variation. The study contributes to property research by linking hybrid forecasting with housing submarket segmentation and by showing how aggregate forecasts can be translated into district-level predictions in a volatile urban market. The framework can support policymakers, urban planners, investors, and financial analysts in monitoring price dynamics, affordability pressures, and investment risk under changing market conditions.