Market Size Forecasting for Embodied Intelligent Robots Using a Transformer–Random Forest Ensemble

Accurate forecasting of the embodied intelligent robot (EIR) market is valuable yet difficult, because its time series are nonlinear, multi-regime, and noisy, driven by fluctuating policy support, uneven regional adoption rates, and rapid iteration of underlying robotics technologies. This paper proposes a hybrid forecasting framework that couples a Transformer encoder with a Random Forest (RF) regressor and fuses their predictions through a weighted ensemble whose combination coefficient is selected by grid search. A comprehensive market dataset spanning 2018–2026 across six regions, eight application scenarios, and six robot categories is analyzed. Lag, rolling statistic, difference, and momentum features are engineered from region-level time series to capture temporal trends and market oscillations. On a held-out 2025–2026 test period, the proposed ensemble attains the lowest MAE (7.76) and MAPE (13.63%), while the standalone Transformer attains the highest R² (0.55). The study demonstrates the complementary strengths of deep sequence modeling and tree-based regression, while documenting the key limitations that need to be addressed before deploying this prediction pipeline in real-world market planning and investment decision-making.