clustering

Recent advances in energy harvesting technology cause Wireless Sensor Networks (WSNs) to be one of the most common communications in different fields. In this paper, a new and appropriate solution based on geographic routing in WSNs with mobile stations is presented to decrease the number of routings and ultimately reduce the usage of energy in the network. In the proposed method named Geographic Routing with the Mobile Sink (GRMS), in the process of selecting the Cluster Head (CH) element, each node that is selected as the CH announces its new role to the other network nodes. By announcing this information, other nodes are also connected to the cluster. In each cluster, the CH creates and broadcasts a schedule in the cluster that contains the time slots assigned to each cluster member. The proposed GRMS method is compared with GEAR, GAF, and GMR methods. The simulation results show that the GRMS has reduced energy consumption compared to GEAR, GAF, and GMR methods. Also, the end-to-end delay in packet delivery has decreased and the number of dead nodes has decreased

Building fixtures like lighting are very important to be modelled, especially when a higher level of modelling details is required for planning indoor renovation. LIDAR is often used to capture these details due to its capability to produce dense information. However, this led to the high amount of data that needs to be processed and requires a specific method, especially to detect lighting fixtures. This work proposed a method named Size Density-Based Spatial Clustering of Applications with Noise (SDBSCAN) to detect the lighting fixtures by calculating the size of the clusters and classifying them by extracting the clusters that belong to lighting fixtures. It works based on Density-Based Spatial Clustering of Applications with Noise (DBSCAN), where geometrical features like size are incorporated to detect and classify these lighting fixtures. The final results of the detected lighting fixtures to the raw point cloud data are validated by using F1-score and IoU to determine the accuracy of the predicted object classification and the positions of the detected fixtures. The results show that the proposed method has successfully detected the lighting fixtures with scores of over 0.9. It is expected that the developed algorithm can be used to detect and classify fixtures from any 3D point cloud data representing buildings.

Various methods have been used for clustering big data. Pattern recognition methods are suitable methods for clustering these data. Due to the large volume of samples taken in the drilling of mines and their analysis for various elements, this category of geochemical data can be considered big data. Examining and evaluating drilling data in the Lar copper mine in Sistan and Baluchistan province located in the southeast of Iran requires the use of these methods. Therefore, the main goal of the article is the clustering of the drilling data in the mentioned mine and its zoning of the geochemical data. To achieve this goal, 3500 samples taken from drilling cores have been used. Elemental analysis for six elements has been done using the ICP-Ms method. Pattern recognition methods including SOM and K-MEANS have been used to evaluate the relation between these elements. The silhouette method has been used to determine and evaluate the number of clusters. Using this method, 4 clusters have been considered for the mentioned data. According to this method, it was found that the accuracy of clustering is higher in the SOM method. By considering the 4 clusters, 4 zones were identified using clustering methods. By comparing the results of the two methods and using the graphical method, it was determined that the SOM method has a better performance for clustering geochemical data in the studied area. Based on that, zones 2 and 4 were recognized as high-grade zones in this area.

The purpose of this study is to design and analyze a model for exit strategies from recession in companies operating in the automotive parts manufacturing industry. This study is of a developmental-applied nature and follows a mixed-methods approach (qualitative and quantitative). The qualitative section is based on a multi-grounded theory approach to develop a qualitative model of dimensions, combining empirical exploratory findings (interviews) and theoretical exploratory findings (meta-synthesis of scientific documents). Quantitative analyses were conducted using the fuzzy cognitive map approach and clustering of companies based on k-means partitioning. The validity of the qualitative findings was confirmed using content validity for constructs and the opinions of external experts. The quantitative research population consisted of two groups: an expert panel of 18 individuals and managers and directors of automotive parts manufacturing companies located in northwest Iran. The expert panel was selected through purposive sampling, while the sample of companies, comprising 360 firms, was selected using stratified random sampling based on provincial distribution. The qualitative findings identified a model of exit strategies from recession for automotive parts manufacturers within a framework of 63 key concepts, 11 subcategories, and 3 major categories. The quantitative findings resulted in the development of a relational model comprising one sender node, two receiver nodes, and eight central nodes. Company clustering identified three major clusters of firms in response to recession exit strategies. The overall analysis of corporate exit strategies from recession indicated that automotive parts manufacturers should adopt a long-term plan focused primarily on transformation-oriented strategies to achieve a suitable level of competitiveness and competitive advantage in the business environment.

Data visualization is a key component of undirected data mining that it transforms data, information, and knowledge into visual view. In this paper, we formulate data visualization problem as a quadratic assignment problem (DV-QAP). The QAP is an NP-Hard problem and has high complexity that it is more acute for data visualization problem because it has intense dependencies among variables and big search space. Therefore, the exact approaches are inefficient to solve DV-QAP and we introduce a new technique called Distributed Self Adaptive Genetic Algorithm with Migration (DSAGAM) that their parameters adjust to increases the exploration and exploitation. This paper focuses on the effect of controlling the migration process and adjusting parameters with respect to the fitness to explore such big search spaces to improve solutions quality. Then we demonstrate the efficiency of the model for a real data set compared with the SGA, SAMGA, IGA and Sammon's mapping approaches.

Nowadays, online shopping plays a vital role in providing services and delivering goods to customers in the context of business intelligence and e-commerce. This research analyzes the customer purchase data of an Iranian online shopping company in Tehran. Among the available datasets provided by the company, 200 thousand records of one week of transactions have been selected for the present study. Several classification methods (i.e., Random Forest, gradient-boosted trees, K-Nearest Neighbor (KNN), Naïve Bayes, Kernel Naïve Bayes, and Neural Networks) and clustering approaches have been applied to discover the knowledge and patterns. The results show that before balancing the dataset, the KNN algorithm with K=5 is the best classification method among the existing methods. However, after balancing, gradient-boosted trees outperform the other classification methods. For clustering methods, the results show that the K-Means algorithm with K=3 is more efficient regarding the average within centroid distance for each cluster. Finally, concluding remarks and suggestions for future studies are stated.

Clustering is a well-known solution to deal with complex database features as an unsupervised machine learning technique. One of its practical applications is the selection of non-similar earthquakes for consequent analysis of structural models. In the present work, appropriate clustering of seismic data is searched via optimization. Silhouette value is penalized and used to define the performance objective. A stochastic search algorithm is combined with a greedy search to solve the problem for distinct sets of near–field and far-field ground motion records. The concept of coherency is borrowed from optics to propose a coherency metric for earthquake signals before and after being filtered by structural models. It is then evaluated for various cases of structural response-to-record and response-to-response comparisons. According to the results the proposed coherency detection procedure performs well; confirmed by distinguished structural response spectra between different clusters.

Iranian Commercial banks are always considered as one of the most important institutions active in the money and capital market, due to the economic structure of the country and the lack of development of the capital markets, which makes them in charge of financing the economic sectors of the country. However, these banks are not successful in fulfilling their mission. High level of banks' reserves shows that they do not pay enough attention to risk management and credit portfolio management. There are several models such as linear programming, integer programming, zero and one programming that can provide an optimal combination of the elements that make up the facility basket. However, entering financial information into mathematical planning by considering all conditions is not straightforward to achieve such a goal. In this research, using data mining to optimize the multi-objective model of facility allocation is done using neural network. First, the effective variables were extracted from the bank database and after preparation, the most important features were identified using different algorithms such as random forest algorithm, MARS, and step-wise regression. Then, these methods were compared with each other and the best method was selected. In order to cluster the customers, k-means and k-medoids models have been used. Using different criteria, including the silhouette and the best number of clusters, two clusters have been estimated and customers have been identified in two low-risk and high-risk categories. And finally, by using convolutional neural network, the risk and profit of each customer has been predicted.