PMKT: Privacy-preserving Multi-party Knowledge Transfer for financial market forecasting

While decision-making task is critical in knowledge transfer, particularly from multi-source domains, existing knowledge transfer approaches are not generally designed to be privacy preserving. This has potential legal and financial implications, particularly in sensitive applications such as financial market forecasting. Therefore, in this paper, we propose a Privacy-preserving Multi-party Knowledge Transfer system (PMKT), based on decision trees, for financial market forecasting. Specifically, in PMKT, we leverage a cryptographic-based model sharing technique to securely outsource knowledge reflected in decision trees of multiple parties, and design a secure computation mechanism to facilitate privacy-preserving knowledge transfer. An encrypted user-submitted request from the target domain can also be sent to the cloud server for secure prediction. Also, the use of decision trees allows us to provide interpretability of the predictions. We then demonstrate how PMKT can achieve privacy guarantees, and empirically show that PMKT achieves accurate forecasting without compromising on accuracy.