Optimized Privacy Preserved Itemset Mining using Federated Learning from Transactional Data in Data Mining

Federated learning allows you to train a global machine-learning model without requiring you to move data from one location to another. This is especially important for applications in the healthcare industry, where data is full of sensitive, personally identifiable data, and data analysis techniques need to demonstrate that they adhere to legal requirements. The created machine learning model or the model variables that are made public during training can still be the target of privacy attacks, even when federated learning forbids the sharing of raw data. In this research, we first present an embedding model for the transaction classification job based on federated learning. Transaction data is viewed by the model as a collection of frequent item sets. After that, by maintaining the contextual relationship between frequent item sets, the algorithm can learn low-dimensionality continuous matrices. We conduct a thorough experimental investigation on a large volume of high-dimensional transactional data to validate the created models that incorporate federated learning and attention-based techniques. Our investigations demonstrate how the categorization might aid in the design of federated learning systems. We provide the design considerations, case cases, and prospects for future research by methodically summarising the current federated learning systems.