The number of devices connected to the Internet will increase exponentially by 2020, which is smoothly migrating the Internet from an Internet of people towards an Internet of Things (IoT). These devices can communicate with each other and exchange information forming a wide Wireless Sensor Network (WSN). WSNs are composed mainly of a large number of small devices that run on batteries, which makes the energy limited. Therefore, it is essential to use an energy efficient routing protocol for WSNs that are scalable and robust in terms of energy consumption and lifetime. Using routing protocols that are based on clustering can be used to solve energy problems. Cluster-based routing protocols provide an efficient approach to reduce the energy consumption of sensor nodes and maximize the network lifetime of WSNs. In this thesis, a single hop cluster-based network layer routing protocol, referred to as HRHP, is designed. It applies centralized and deterministic approaches for the selection of cluster heads, in relation to offer an improved network lifetime for large-scaled and dense WSN deployments. The deterministic approach for selecting CHs is based on the positive selection mechanism in the human thymus cells (T-cells). HRHP was tested over six different scenarios with BS position outer the sensing area, it achieved a maximum average of 78% in terms of life time. To further reduce energy consumption in WSN, a multi-hop algorithm, referred to as MLHP, is proposed for prolonging the lifetime of WSN. In this algorithm, the sensing area is divided into three levels to reduce the communication cost by reducing the transmission distances for both inter-cluster and intra-cluster communication. MLHP was tested over fourteen cases with different heterogeneity factors and area sizes and achieved a maximum of 80% improvement in terms of life time. Finally, a real-time and autonomous emergency evacuation approach is proposed, referred to as ARTC-WSN, which integrates cloud computing with WSN in order to improve evacuation accuracy and efficiency for smart buildings. The approach is designed to perform localized, autonomous navigation by calculating the best evacuation paths in a distributed manner using two types of sensor nodes (SNs), a sensing node and a decision node. ARTC-WSN was tested in five scenarios with different hazard intensity, occupation ratio and exit availability over three different areas of evacuation and achieved an average of 98% survival ratio for different cases.

Advances in technology have provided industry with an array of devices for collecting data. The frequency and scale of data collection means that there are now many large datasets being generated. To find patterns in these datasets it would be useful to be able to apply modern methods of classification such as support vector machines. Unfortunately these methods are computationally expensive, quadratic in the number of data points in fact, so cannot be applied directly. This thesis proposes a framework whereby a variety of clustering methods can be used to summarise datasets, that is, reduce them to a smaller but still representative dataset so that these advanced methods can be applied. It compares the results of using this framework against using random selection on a large number of classification and regression problems. Results show that the clustered datasets are on average fifty percent smaller than the original datasets without loss of classification accuracy which is significantly better than random selection. They also show that there is no free lunch, for each dataset it is important to choose a clustering method carefully.

<p>This thesis covers the analysis of sparse and irregular fuel consumption data of long</p><p>distance haulage articulate trucks. It is shown that this kind of data is hard to analyse with multivariate as well as with functional methods. To be able to analyse the data, Principal Components Analysis through Conditional Expectation (PACE) is used, which enables the use of observations from many trucks to compensate for the sparsity of observations in order to get continuous results. The principal component scores generated by PACE, can then be used to get rough estimates of the trajectories for single trucks as well as to detect outliers. The data centric approach of PACE is very useful to enable functional analysis of sparse and irregular data. Functional analysis is desirable for this data to sidestep feature extraction and enabling a more natural view on the data.</p>

We demonstrate the application of CMB techniques to measure the clustering of infrared emitting star-forming galaxies. We detect correlations in the cosmic far-infrared background due to the clustering of star-forming galaxies in observations made with the Balloon-borne Large Aperture Submillimeter Telescope, BLAST, at 250, 350, and 500μm. We perform jackknife and other tests to confirm the reality of the signal. The measured correlations are well fit by a power law over scales of 5–25 arcminutes, with ∆I/I = 15.1 ± 1.7%. We adopt a specific model for submillimeter sources in which the contribution to clustering comes from sources in the redshift ranges 1.3≤z≤2.2, 1.5≤z≤2.7,and1.7≤z≤3.2,at 250, 350 and 500 μm, respectively. With these distributions, our measurement of the power spectrum, P(kθ), corresponds to linear bias parameters, b = 3.8±0.6,3.9±0.6 and 4.4±0.7, respectively. We further interpret the results in terms of the halo model, and find that at the smaller scales, the simplest halo model fails to fit our results. One way to improve the fit is to increase the radius at which dark matter halos are artificially truncated in the model, which is equivalent to having some star-forming galaxies at z ≥ 1 located in the outskirts of groups and clusters. In the context of this model we find a minimum halo mass required to host a galaxy is log(Mmin/M⊙) = 11.5+0.4, and we derive effective biases beff = 2.2 ± 0.2, 2.4 ± 0.2, −0.1 and 2.6 ± 0.2, and effective masses log(Meff/M⊙) = 12.9 ± 0.3, 12.8 ± 0.2, and 12.7 ± 0.2 , at 250, 350 and 500 μm, corresponding to spatial correlation lengths of r0 = 4.9, 5.0, and 5.2 ±0.7 h−1 Mpc, respectively. Finally, we discuss implications for clustering measurement strategies with Herschel and Planck.

This thesis proposes a clustering algorithm for vehicular communication networks. A novel clustering metric and an improved clustering framework are introduced. The novel clustering metric, network criticality, is a global metric on undirected graphs which quantifies the robustness of the graph against changes in environmental parameters, and point-to-point network criticality is also defined to measure the resistance between different points of a graph. We localize the notion of network criticality for a node of a vehicular network which can potentially be promoted as the cluster header. We use the localized notion of node criticality in conjunction with a universal link metric, Link Expiration Time (LET), to derive a clustering algorithm for the vehicular network. We employ a distributed multi-hop clustering algorithm based on the notion of network criticality. Simulation results show that the proposed clustering algorithm forms a more robust cluster structure.

This thesis proposes a clustering algorithm for vehicular communication networks. A novel clustering metric and an improved clustering framework are introduced. The novel clustering metric, network criticality, is a global metric on undirected graphs which quantifies the robustness of the graph against changes in environmental parameters, and point-to-point network criticality is also defined to measure the resistance between different points of a graph. We localize the notion of network criticality for a node of a vehicular network which can potentially be promoted as the cluster header. We use the localized notion of node criticality in conjunction with a universal link metric, Link Expiration Time (LET), to derive a clustering algorithm for the vehicular network. We employ a distributed multi-hop clustering algorithm based on the notion of network criticality. Simulation results show that the proposed clustering algorithm forms a more robust cluster structure.

We demonstrate the application of CMB techniques to measure the clustering of infrared emitting star-forming galaxies. We detect correlations in the cosmic far-infrared background due to the clustering of star-forming galaxies in observations made with the Balloon-borne Large Aperture Submillimeter Telescope, BLAST, at 250, 350, and 500μm. We perform jackknife and other tests to confirm the reality of the signal. The measured correlations are well fit by a power law over scales of 5–25 arcminutes, with ∆I/I = 15.1 ± 1.7%. We adopt a specific model for submillimeter sources in which the contribution to clustering comes from sources in the redshift ranges 1.3≤z≤2.2, 1.5≤z≤2.7,and1.7≤z≤3.2,at 250, 350 and 500 μm, respectively. With these distributions, our measurement of the power spectrum, P(kθ), corresponds to linear bias parameters, b = 3.8±0.6,3.9±0.6 and 4.4±0.7, respectively. We further interpret the results in terms of the halo model, and find that at the smaller scales, the simplest halo model fails to fit our results. One way to improve the fit is to increase the radius at which dark matter halos are artificially truncated in the model, which is equivalent to having some star-forming galaxies at z ≥ 1 located in the outskirts of groups and clusters. In the context of this model we find a minimum halo mass required to host a galaxy is log(Mmin/M⊙) = 11.5+0.4, and we derive effective biases beff = 2.2 ± 0.2, 2.4 ± 0.2, −0.1 and 2.6 ± 0.2, and effective masses log(Meff/M⊙) = 12.9 ± 0.3, 12.8 ± 0.2, and 12.7 ± 0.2 , at 250, 350 and 500 μm, corresponding to spatial correlation lengths of r0 = 4.9, 5.0, and 5.2 ±0.7 h−1 Mpc, respectively. Finally, we discuss implications for clustering measurement strategies with Herschel and Planck.

Clustering, which is partitioning data into groups of similar objects, has a wide range of applications. In many cases unstructured data makes up a significant part of the input. Attempting to cluster such part of the data, which can be referred to as noise, can disturb the clustering on the remaining domain points. Despite the practical need for a framework of clustering that allows a portion of the data to remain unclustered, little research has been done so far in that direction. In this thesis, we take a step towards addressing the issue of clustering in the presence of noise in two parts. First, we develop a platform for clustering that has a cluster devoted to the "noise" points. Second, we examine the problem of "robustness" of clustering algorithms to the addition of noise. In the first part, we develop a formal framework for clustering that has a designated noise cluster. We formalize intuitively desirable input-output properties of clustering algorithms that have a noise cluster. We review some previously known algorithms, introduce new algorithms for this setting, and examine them with respect to the introduced properties. In the second part, we address the problem of robustness of clustering algorithms to the addition of unstructured data. We propose a simple and efficient method to turn any centroid-based clustering algorithm into a noise robust one that has a noise cluster. We discuss several rigorous measures of robustness and prove performance guarantees for our method with respect to these measures under the assumption that the noise-free data satisfies some niceness properties and the noise satisfies some mildness properties. We also prove that more straightforward ways of adding robustness to clustering algorithms fail to achieve the above mentioned guarantees.

With the proliferation of electronic commerce and knowledge economy environments, both persons and organizations increasingly have generated and consumed large amounts of online information, typically available as textual documents. To manage this rapid growth of the number of textual documents, people often use categories or folders to organize their documents. These document grouping behaviors are intentional acts that reflect the persons¡¦ (or organizations¡¦) preferences with regard to semantic coherency, or relevant groupings between subjects. For this thesis, we design and implement an ontology-based personalized document clustering (OnPEC) technique by incorporating both an individual user¡¦s partial clustering and an ontology into the document clustering process. Our use of a target user¡¦s partial clustering supports the personalization of document categorization, whereas our use of the ontology turns document clustering from a feature-based to a concept-based approach. In addition, we combine two hierarchical agglomerative clustering (HAC) approaches (i.e., pre-cluster-based and atomic-based) in our proposed OnPEC technique. Using the clustering effectiveness achieved by a traditional content-based document clustering technique and previously proposed feature-based document clustering (PEC) techniques as performance benchmarks, we find that use of partial clusters improves document clustering effectiveness, as measured by cluster precision and cluster recall. Moreover, for both OnPEC and PEC techniques, the clustering effectiveness of pre-cluster-based HAC methods greatly outperforms that of atomic-based HAC methods.

According to the context theory of classification, the document-clustering behaviors of individuals not only involve the attributes (including contents) of documents but also depend on who is doing the task and in what context. Thus, effective document-clustering techniques need to be able to take into account users¡¦ categorization preferences and thus can generate document clusters from different preferential perspectives. The Preference-Anchored Document Clustering (PAC) technique was proposed for supporting preference-based document-clustering. Specifically, PAC takes a user¡¦s categorization preference into consideration and subsequently generates a set of document clusters from this specific preferential perspective. In this study, we attempt to investigate two research questions concerning the PAC technique. The first research question investigates ¡§whether the incorporation of the broader-term expansion (i.e., the proposed PAC2 technique in this study) will improve the effectiveness of preference-based document-clustering, whereas the second research question is ¡§whether the use of a statistical-based thesaurus constructed from a larger document corpus will improve the effectiveness of preference-based document-clustering.¡¨ Compared with the effectiveness achieved by PAC, our empirical results show that the proposed PAC2 technique neither improves nor deteriorates the effectiveness of preference-based document-clustering when the complete set of anchoring terms is used. However, when only a partial set of anchoring terms is provided, PAC2 cannot improve and even deteriorate the effectiveness of preference-based document-clustering. As to the second research question, our empirical results suggest the use of a statistical-based thesaurus constructed from a larger document corpus (i.e., the ACM corpus consisting of 14,729 documents) does not improve the effectiveness of PAC and PAC2 for preference-based document-clustering.