A Perceptually Important Points Approach Based on Imputation Clustering with Weighted Distance Techniques for Big Data Reduction in Internet of Things Cloud

IoT sensing devices tend to generate large volume of data samples consisting of relevant and irrelevant sensed data records. Irrelevant data points are regarded as data redundancy which are mainly eliminated from the data samples to achieve relevant ones for onward processing. Several researches have made significant effort to detect and eliminate redundant sensing data points with the support of dimensionality reduction techniques. These techniques mainly remove redundant data records by similarity comparison between the features of a given sensed dataset without considering the data records or points. However, there is a technique called Perceptually Important Point (PIP) deployed to eliminate data redundancy that considers the sensed data records but proved ineffective as it eliminates relevant sensed data alongside with redundant ones due to missing data. Therefore, K-means imputation clustering with the combination of Cosine and Manhattan Weighted Distance Measure technique is proposed in this research. Thus, for the recovery of missing data in order to improve the performance of the PIP technique. Simulations are conducted on five benchmark datasets for the elimination of redundant sensed data records. Experimental results shows that the proposed model outperforms the existing PIP technique with up to 99.967 and 99.614% accuracy with the execution time of 1234 s, before and after the elimination of redundant sensed data records on a given datasets.