Dimensionality Reduction Techniques for IoT Based Data

Abstract

Background: Internet of Things (IoT) plays a vital role by connecting several heterogeneous devices seamlessly via the Internet through new services. Every second, the scale of IoT keeps on increasing in various sectors like smart home, smart city, health, smart transportation and so on. Therefore, IoT becomes the reason for the massive rise in the volume of data which is computationally difficult to work out on such a huge amount of heterogeneous data. This high dimensionality in data has become a challenge for data mining and machine learning. Hence, with respect to efficiency and effectiveness, dimensionality reduction techniques show the roadmap to resolve this issue by removing redundant, irrelevant and noisy data, making the learning process faster with respect to computation time and accuracy. Methods: In this study, we provide a broad overview on advanced dimensionality reduction techniques to facilitate selection of required features necessary for IoT based data analytics and for machine learning on the basis of criterion measure, training dataset and inspired by soft computation technology followed by significant challenges of dimensionality reduction techniques for IoT generated data that exists as scalability, streaming datasets and features, stability and sustainability. Results and Conclusion: In this survey, the various dimensionality reduction algorithms reviewed delivers the essential information in order to recommend the future prospect to resolve the current challenges in the use of dimensionality reduction techniques for IoT data. In addition, we highlight the comparative study of various methods and algorithms with respect to certain factors along with their pros and cons.