Abstract: Amidst the accelerating global energy transition, cobalt has emerged as a critical strategic mineral indispensable for power batteries and high-performance alloys, resulting in a sustained surge in its demand. Conventional cobalt ore prospecting prediction methods, which heavily depend on the empirical expertise of geologists and the integrative interpretation of multi-source geoscientific data, are not only time-consuming and labor-intensive but also inherently subjective. Currently, artificial intelligence, particularly deep learning, is profoundly transforming the field of mineral exploration and prospecting. In this study, an intelligent ore prospecting prediction model based on a multi-branch convolutional neural network （MB−CNN） was developed to enable efficient and accurate prediction of cobalt mineralization potential. The model innovatively designed a parallel artificial neural network architecture. It extracted deep features from heterogeneous data sources, including geological, geophysical, and geochemical data, through three distinct convolutional neural networks and subsequently integrated these features. The trained model was applied to the Zhongwei Area in Ningxia. The experimental results demonstrate that compared with the weight of evidence method and CNN2D model, the constructed intelligent ore prospecting prediction model shows significant improvements in both prediction accuracy and robustness. Specifically, compared to CNN2D, the prediction accuracy of MB−CNN is increased by 9.39 percentage point, while the prediction area proportion is reduced by 7.73 percentage point, effectively lowering the uncertainty of the prediction results. This provides a powerful intelligent solution for achieving breakthroughs in cobalt ore prospecting in new areas.