Intelligent Mortgage Rate Forecasting Using AI: Enhancing Accuracy, Governance, and Financial System Resilience

Abstract

The mortgage rate forecasting is a difficult task because financial markets are subject to complex interaction of macroeconomic
indicators and market dynamics. In this paper, an intelligent AI-based prediction system in mortgage rates prediction is proposed
based on Multisource Stock Market Trends dataset that involves stock prices, economic variables, and sentiment variables. The
processing of the data is done via the treatment of the missing values, encoder of the sentiment, feature engineering (lag, rolling,
and interaction features), StandardScaler scaling, and the time order division of the data. In order to achieve better predictive
performance, a hybrid ensemble model is resorted to that is trained with XGBoost and LightGBM as well as a meta-learner
which is stacking-based. The experimental findings which show that the hybrid model is superior to the base models since it
offers a superior accuracy with a higher value of R2 of 0.9799, lower value of RMSE of 0.1920, lower value of MAE of 0.1462
and less value of M APE of 2.69%. Moreover, the study employs explainable AI (SHAP), stress testing, and stability testing with
the aim of ensuring transparency and strength. The findings reveal how the proposed framework can be useful in improving
the precision of the prediction and facilitating the implementation of the model governance and resilience evaluation AI-based
mortgage rate forecasting.