Adaptive Regime-Aware Stock Price Prediction Using Autoencoder-Gated Dual Node Transformers with Reinforcement Learning Control
Mohammad Al Ridhawi, Mahtab Haj Ali, Hussein Al Osman
This paper proposes an adaptive stock price prediction system that automatically detects different market conditions (stable vs. volatile) and routes data through specialized neural networks designed for each regime, rather than using a one-size-fits-all approach. The system uses an autoencoder to identify unusual market conditions, dual transformer networks to handle different market states, and reinforcement learning to automatically adjust how it switches between prediction pathways based on performance. Tested on 20 major stocks from 1982-2025, the full system achieves 0.59% prediction error (vs. 0.80% for simpler models) and maintains accuracy even during volatile market crashes.
time series forecastingstock price predictiontransformersautoencoders