Optimizing Industrial IoT Systems with Multimodal Federated Learning: Challenges and Solutions

With the emergence of many loT devices to cater to human necessities, challenges exist in the form of privacy, challenging communication protocols, and data heterogeneity arising from the generation of multimodal data. To address these challenges, this research studies a novel Multimodal Federated Learning (MFL) approach. MFL enables the distributed training of machine learning models on IoT devices so that data security is achieved and use is made of the variety of data available. Compared to the other experimental results, it is evident that the proposed MFL approach achieves better results than the FL and centralized methods. The performance of all four indices, the accuracy of the model, the precision, the recall, and the F1-score, all were enhanced from a 10% baseline of 82% to 90% after 10 iterations of the algorithm. These variables were obtained from the correlation study with values greater than 0 for all, indicating strong positive associations—97, indicating the efficiency improvement with a significant difference.Furthermore, along with the principles of MFL, it has been observed that the amount of communication overhead has been reduced from as much as 50 MB to as low as 26 MB, and the amount of training time has been reduced from 120 seconds to 93 seconds which is quite efficient and can be implemented on a large scale. Concerning the revealed general findings, the MFL technique’s enhancement in integrating various types of data to enhance predictive analytics, resource utilization, and decision-making within the IoT systems is recognized.