الفهرس 
Chapter1: IntroductionOnly 14 pages are availabe for public view
 |  | from | 255 |  |  |
| | | from | 255 | | |
Abstract
Internet of Things (IoT) has penetrating all things around us to give them the ability to produce piece of information about their states on the Internet. As a result, Smart Things (SThs) produce massive real-time data (i.e., big IoT data). In general, resulting deluge of real-time data streams has the property of five V of the big data (i.e., volume, velocity, variety, veracity, and value). The potential of such real-time data is to be retrieved, mined, and analyzed in the real-time to empower the future IoT application. Consumers of IoT data are not only human users, but also other SThs in IoT applications. This cooperation between SThs for enabling data sharing reveals the paradigm of Industrial IoT (IIoT). Smartness of IoT applications relies on automatic control, events handling, and decision making. For enabling these features, search and discovery services have to be enhanced in IoT application.
The main problem concerning searching the IoT concentrates on three main parts (a) how to keep indexes as up-to-date as possible, (b) how to index such deluge of data streams, and (c) how to crawl SThs in a dynamic network of networks (i.e., IoT applications). Recent research studies as declared in related work sections present multiple solutions for handling big data in the IoT, for example, Dyser and WoTSF propose indexing prediction models for SThs. Other works propose building range queries for enabling IoT search engines to capture the intended SThs, on the other hand, other research works presents pattern query search frameworks such as similar sensor search. Based on these works, this dissertation presents a promising solution to reduce redundancy data, which are used for increasing sensing accuracy by summarizing similar time series capturing most of SThs behavior in one time series to be indexed as a cluster representative in the higher level indexes as proposed in WoTSF.
Main contributions of this thesis are listed in brief:
• Presenting a comprehensive review for classifying Internet of Things (IoT) challenges and presenting recommended and promising technologies that could address most of those challenges.
• Presenting novel balanced indexing models for IoT resources, the proposed algorithms of these models base on Dynamic Time Warping (DTW) for achieving similarity data fusion for IoT resources. These algorithms are called time series Cluster Representatives (ClRe) for time series clusters. Based on the experimental results, ClRe reduces redundancy by O(K-1), where K is number of time series in a cluster.
• Also this thesis presents modified versions of ClRe called ClRe extensions for balancing indexing, achieving high accuracy for the indexed data, and for achieving multiple goals such as (best running time, best accuracy, and best reduced representative).
• Novel execution methods are proposed to improve performance of all ClRe extensions, these methods are called Linear and Pair-merge execution methods.
• In addition, static and dynamic communication models are presented for reducing data transmission to increase SThs life-time.
• Moreover, this method enables building smart schedule for crawling SThs in IoT applications for keeping indexes as up-to-date as possible.
Evaluation for ClRe algorithms and their extensions were done using real examples and using real dataset (Szeged-weather). Results proves that ClRe 3.0 has achieve balancing indexing with higher accuracy (≈ 95%) and reduces indexes for similar SThs by O(K-1), its extension ClRe 3.1 increases its accuracy with slight or negligent increase in running time and indexes sizes. This extension captures most of SThs behaviors to increase similarity rate by ≈ 9 %. On the other hand, ClRe 1.1 achieves the better running time, where it relies FastDTW for selecting the best time series to be the cluster representative. For indexing only common behaviors of SThs, ClRe 3.2 generates less length representatives. By running ClRe extensions especially extensions of ClRe 3.0, pair-merge produces more accurate results ≈ 21.69%.