Abstract:
Internet of Things (IoT), being a global infrastructure, is expected to improve the quality of life. It has a potential to bring intelligent solutions in many different domains such as in the health industry, smart homes, transportation systems etc. The potential applicability of IoT to almost every field of life gives rise to an increased adoption of IoT solutions across the world. The increase in the number of IoT solutions and IoT devices in turn, leads to generation of large amount of data. Collecting, communicating, storing and extracting useful information from this huge amount of data is a challenging task. Some effective methods are therefore required to handle and analyze IoT data, so that some meaningful information can be extracted out of it. Dimension reduction is one such method which helps to retain only the meaningful features of the data by removing the irrelevant ones, thus helping to have reduced data with only the necessary information. This study focuses on how dimension reduction can be used to improve overall costs of IoT solutions by optimizing the communication and storage requirements. It aims to identify the influence of dimension reduction on IoT data in terms of storage and communication cost. The effect of dimensionality reduction on the classification performance of different classifiers has also been analyzed. This analysis investigates how dimension reduction may affect different data mining tasks that might be carried out on the IoT data stored on to the cloud. The analysis is carried out by applying a few dimensionality reduction techniques on two IoT based datasets. It has been found that dimension reduction helps to reduce the storage and communication costs of IoT data. Furthermore, it is observed that the performance of different classification algorithms reduced to a small extent, when these algorithms are applied to dimensionally reduced data. However, this reduction in performance is negligible as compared to the optimized storage and communication cost achieved by applying dimension reduction on data generated by IoT systems.
