On Explaining Federated Learning Models for Classification Tasks