Abstract
Predicting price spikes in critical metals such as Cobalt, Copper, Magnesium,and Nickel is crucial for mitigating economic risks associated with globaltrends like the energy transition and reshoring of manufacturing. Whiletraditional models have focused on regression-based approaches, our workintroduces a neurosymbolic ensemble framework that integrates multiple neuralmodels with symbolic error detection and correction rules. This framework isdesigned to enhance predictive accuracy by correcting individual model errorsand offering interpretability through rule-based explanations. We show that ourmethod provides up to 6.42% improvement in precision, 29.41% increase in recallat 13.24% increase in F1 over the best performing neural models. Further, ourmethod, as it is based on logical rules, has the benefit of affording anexplanation as to which combination of neural models directly contribute to agiven prediction.