Conventional Belief Rule Base (BRB) models excel at transparent reasoning under uncertainty but struggle when the underlying data distribution drifts. We tackle this limitation by injecting momentum awareness into the Extended BRB (EBRB). First, we devise the Momentum Differential Index (MDI), a light-weight statistic that tracks first- and higher-order rate-of-change in any temporal feature. Next, we embed this signal into a new Dynamic Momentum-Driven Weight Optimisation (DMWO) algorithm that rescales gradient updates with MDI-weighted velocity terms while preserving the \(\Theta (mn)\) complexity of classic EBRB training. Across a five-season European-football corpus DMWO (i) raises macro-F1 by +2.2 pp, (ii) lowers log-loss by 13 %, (iii) cuts expected-calibration-error by 23 %, and (iv) limits catastrophic forgetting to 2.3 % in a five-year online deployment-each improvement being statistically significant. Cross-domain tests on the Kaggle Stocks Daily, CWRU Bearing, and UCI Air Quality benchmarks confirm consistent accuracy gains (+4.3−4.7 pp) with the lowest prediction variance. Despite the added momentum terms, the framework sustains 20 k predictions s−1 on a single RTX-3060 GPU, incurring only an 11 % runtime overhead and a 0.3 GB memory increase. Crucially, rule-level reasoning traces show that DMWO adjusts numeric belief degrees, not linguistic antecedents, thereby retaining full interpretability. The proposed MDI-EBRB architecture therefore delivers a rare combination of adaptability, scalability, and transparency, making it a strong candidate for high-velocity, non-stationary decision-support applications in sports analytics, finance, industrial monitoring, and environmental sensing.
