Federated Learning (FL) has emerged as an efficient technique to train machine learning (ML) models across decentralized devices without sharing any raw data for preserving privacy. However, it faces challenges in communication overhead and resource constraints, particularly in real-time applications of Internet of Things (IoT). To address these challenges, we propose a novel method called Hashcash and Jaya-based communication efficient Federated Learning (HJCFL). It consists of three main phases: local aggregator selection, clustering of IoT devices and communication efficient FL process. It uses Hashcash for strategically selecting local aggregators and Jaya based algorithm for clustering IoT devices. The usage of Hashcash function has an added advantage over existing aggregator selection techniques because of it’s simplicity, proof-of-work and adjustable difficulty level. To the best of our knowledge, none of the existing techniques uses Hashcash for the same. Moreover, unlike most of the existing schemes, the proposed method utilizes all the available devices and therefore it converges faster with significantly less number of global rounds than others. Through extensive simulations, we demonstrate the effectiveness of the proposed HJCFL as compared to two baseline techniques. The result’s validity is further confirmed using a widely recognized statistical technique, i.e., analysis of variance (ANOVA), followed by a least significant difference (LSD) post-hoc analysis.