Temporal and spatial data redundancy reduction using machine learning approach for IoT based heterogeneous wireless sensor networks

Data redundancy reduction is one of the efficient power conservation methods in IoT based Wireless Sensor Networks (IoT-WSN). Generally, WSN nodes undergo redundant data transmission both spatially and temporally, which in turn consumes more energy. It also leads to unbalanced big data storage, hence reduces the performance of the database management system. Communicating the aggregate representation of sensed data solves the problems of storage space and the high energy consumption for transmission from the source node to the destination node in the network. In the proposed method, the temporal redundant data is annihilated at the member node level using the Cosine similarity function, the spatial redundant data is obliterated at the cluster head level by the Extreme Learning Machine (ELM) and the data prediction is performed at base station using Long Short-Term Memory networks (LSTM). The proposed Temporal-Spatial Redundancy Reduction and Prediction Algorithm (TSRRPA) aids in enhancing the efficient data transmission and energy utilization in IoT based WSN. This paper provides a detailed discussion about the algorithms at node level, cluster head level, and the base station level in the context of redundancy reduction and the experimental results of the simulation. The TSRRPA exhibits a 38% increase in energy consumption efficiency, also achieves an overall accuracy of 98.7% with an impressively low error rate of 0.013.