clustering

In analyzing phenomena around us, clustering is among the most commonly used techniques in machine learning for comparing, and categorizing them into different groups based on intrinsic features. One of the main challenges facing clustering algorithms is selecting a suitable representative for each cluster. Existing algorithms often choose a single representative, which can lead to suboptimal performance on many datasets (especially asymmetric datasets). This process is completely dependent on the type of internal distribution of the clusters, and that single point may not be a suitable representative for that cluster. The proposed algorithm for dealing with datasets, inspired by the fuzzy ALM method and avoiding complex formulas, and calculations, initially breaks the system down into simpler (two-dimensional) systems. After spreading ink drops, by finding the vertical Narrow path and the horizontal narrow path, it selects a set of points as the representation of each cluster. The proposed algorithm, unlike many conventional algorithms, provides a representative set for each cluster and also enhances the algorithm's performance in dealing with datasets that have an asymmetric structure by introducing a new distance measure based on the KNN method and utilizing the set of prime numbers. The Accuracy, F1-Score, and AMI achieved when working with many low-dimensional, and high-dimensional datasets has been higher compared to algorithms such as FUALM, HiDUALM, K-Means, DBSCAN, DENCLUE and IRFLLRR and in some cases, the achieved accuracy has been equal to 100 percent.

The expansion of machine learning applications across different domains has given rise to a growing interest in tapping into the vast reserves of data generated by edge devices. To preserve data privacy, federated learning was developed as a collaboratively decentralized privacy-preserving technology to overcome the challenges of data silos and data sensibility. This technology faces certain limitations due to the limited network connectivity of mobile devices and malicious attackers. In addition, data samples across all devices are typically not independent and identically distributed, which presents additional challenges to achieve convergence in fewer communication rounds. In this paper, we have simulated attacks, namely Byzantine, label flipping, and noisy data attacks, besides non-IID data. We proposed  Robust federated learning against poisoning attacks (RFCaps) to increase security and accelerate convergence. RFCaps incorporate a prediction-based clustering and a gradient quality evaluation method to prevent attackers from the aggregation phase by applying multiple filters and accelerating convergence using the highest quality gradients. Compared to MKRUM, COMED, TMean, and FedAvg algorithms, RFCap has high robustness in the presence of attackers and has achieved a higher accuracy of up to 80%  on both MNIST and Fashion-MNIST datasets.

The Covid-19 pandemic has caused changes in the field of business activities. The fear of the pandemic has increased consumer awareness of the economic benefits of e-commerce platforms and the desire of people to absentee shopping, resulting in the growth of e-commerce in the world. Iran is not an exception to this rule. In this article, in order to analyze the behavior of e-commerce consumers, before and during covid-19, Iran's largest online retail site has been studied. In the analysis of consumers' behavior, parameters such as spatial distribution, gender and age range of consumers, product groups, payment method for orders according to the amount of use of information and communication technology in the country before and during covid-19 and the death rate due to corona, were considered. In this research, data mining method has been used to analyze the behavior of e-commerce consumers. In the procedure, first the required data is collected and then in the next step, the data has been refined with the KNIME Analytics platform, and then consumer behavior has been analyzed using k-means and Fuzzy c-means algorithms. The results of the analysis showed that the spread of covid-19 has made the amount of shopping at the country level more uniform, leading older people to shop online and increasing their share of the total e-commerce consumers and the desire to pay for orders online. Also, the examination of product groups showed that the sales growth of health product groups was higher than other groups.

Wireless sensor networks contain of many sensors that can serve as powerful tools for data collection in environments. A key challenge in these networks is the limited lifetime of sensor batteries. Ideally, all nodes would exhaust their energy simultaneously or through regular scheduling, maximizing the lifetime. Consequently, the primary concern is achieving optimal energy utilization to extend the network's lifetime over a logical duration. Depleting the batteries of the sensors means stopping the operation of the network, because it is practically impossible to replace the batteries of thousands of nodes. To address this issue, the low energy adaptive clustering hierarchical (LEACH) protocol has been widely recognized as one of the prominent solutions for clustering WSNs. However, the random selection of cluster heads in each round under the LEACH protocol fails to guarantee proper convergence. To overcome this limitation, this paper proposes a refined approach by utilizing a genetic algorithm and a novel objective function that incorporates various factors, including energy level and distance. The algorithm employs chromosomes to represent CHs and facilitates the selection of cluster nodes. Notably, the proposed algorithm dynamically performs clustering, meaning that clustering is conducted iteratively, considering identifying dead nodes. By leveraging this approach, the algorithm significantly enhances the clustering quality, ultimately leading to an increased network lifetime. To validate its effectiveness, it is compared with LEACH, LEACH_E and LEACH_EX algorithms, demonstrating its superior capabilities. On average, the proposed algorithm has more alive nodes in the network, and the remaining energy is at least 11% higher than the best other algorithms.

This paper presents a novel channel model based on the edge detection of the interacting objects (IOs) for massive multiple-input and multiple-output (MIMO) systems considering different propagation phenomena such as reflection, refraction, diffraction, and scattering occur at the edges of the IOs. This channel model also uses cluster concepts to model multipath components (MPCs). The time-variant condition of the channel as well as the Doppler effect is considered in this channel model. Also, the non-stationary property of the channel across the array axis can be observed in the simulations. Due to a large number of antenna elements utilized in the massive MIMO systems, the spherical wavefront is assumed instead of the plane wavefront which is used in the conventional MIMO channel model. The central limit theory is also utilized to model the MPCs of each cluster. Furthermore, specific channel characteristics such as channel impulse response (CIR), angle of arrival (AoA) as well as time of arrival (ToA) are extracted in the simulations of the channel.

In this paper, a new feature extraction method is presented based on spectro-temporal representation of speech signal for phoneme classification. In the proposed method, an artificial neural network approach is used to cluster spectro-temporal domain. Self-organizing map artificial neural network (SOM) was applied to clustering of features space. Scale, rate and frequency were used as spatial information of each point and the magnitude component was used as similarity attribute in clustering algorithm. Three mechanisms were considered to select attributes in spectro-temporal features space. Spatial information of clusters, the magnitude component of samples in spectro-temporal domain and the average of the amplitude components of each cluster points were considered as secondary features. The proposed features vectors were used for phonemes classification. The results demonstrate that a significant improvement is obtained in classification rate of different sets of phonemes in comparison to previous clustering-based methods. The obtained results of new features indicate the system error is compensated in all vowels and consonants subsets in compare to weighted K-means clustering.

In this paper, a new hybrid routing protocol is presented for low power Wireless Sensor Networks (WSNs). The new system uses an integrated piezoelectric energy harvester to increase the network lifetime. Power dissipation is one of the most important factors affecting lifetime of a WSN. An innovative cluster head selection technique using Cuckoo optimization algorithm has been used in the designed protocol. The residual energy of the nodes and the distances to the sink were used in the threshold calculations, besides to take advantage of the relay node for communication. A hybrid method using the optimized routing protocol and the integrated energy harvester results in 100% increase in the network lifetime compared to recent clustering-based protocols. The simulations results using MATLAB indicate that energy consumption was been decreased by more than 40%.

Multivariate time series (MTS) data are ubiquitous in science and daily life, and how to measure their similarity is a core part of MTS analyzing process. Many of the research efforts in this context have focused on proposing novel similarity measures for the underlying data. However, with the countless techniques to estimate similarity between MTS, this field suffers from a lack of comparative studies using quantitative and large scale evaluations. In order to provide a comprehensive validation, an extensive evaluation of similarity measures for MTS clustering were conducted. The 14 well-known similarity measures with their variants and testing their effectiveness on 23 MTS datasets coming from a wide variety of application domains were re-implemented. In this paper, an overview of these different techniques is given and the empirical comparison regarding their effectiveness based on agglomerative clustering task is presented. Furthermore, the statistical significance tests were used to derive meaningful conclusions. It has been found that all of similarity measures are equivalent, in terms of clustering F-measure, and there is no significant difference between similarity measures based on our datasets. The results provide a comparative background between similarity measures to find the most proper method in terms of performance and computation time in this field.

In this paper, a proposed ensemble classification for fully polarimetric synthetic aperture radar (PolSAR) data using a global-local classification approach is presented. In the first step, to perform the global classification, the training feature space is divided into a specified number of clusters. In the next step to carry out the local classification over each of these clusters, which contains elements of several classes, a base classifier is trained. Thus, an ensemble of classifiers has been formed which each of them acts professionally in a part of the feature space. To achieve more diversity, the data set is independently partitioned into variable number of clusters by classifier and K-means algorithm. To combine outputs of different arrangements, majority voting, Naïve Bayes and a heuristic combination rule with taking into account the classification accuracy and reliability (which in PolSAR classification less attention has been paid to it) as objective functions, are used. The experimental results over two PolSAR images prove effectiveness of the proposed algorithms in comparison to the baseline methods.

Due to the extensive use of composites in various industries, and the fact that defects reduce ultimate strength and efficiency during operation, detection of failures in composite parts is very important. The aim of this paper is to use Acoustic Emission (AE) non-destructive method in four-point bending test of carbon/epoxy composite to analyze and examine the failure mechanisms. This method is based on waves activated from defects in structures which are built during loading. Sensors collect acoustic signals which created by the separated layers. Each stage of failure is in specific frequency range and indicates a specific mechanism. Clustering of these signals is done with C-means and K-means algorithms then compared with the results of previous works. The failure process was shown to proceed through four stages. In both algorithms, each cluster coincides with one part of these stages.

The k-means algorithm is known to have a quadratic time complexity in terms of n, the input data size. This quadratic complexity debars the algorithm from being effectively used in large applications. In this article, an attempt is made to develop a faster version (in terms of both number of iterations and execution time) of the classical k-means algorithm which requires comparatively lesser number of iterations. The underlying modification includes a gradual directional movement of intermediate clusters and thereby improving compactness and separability properties of cluster structures simultaneously. This process also results in an improved visualization of clustered data. Experimental results using various types of data sets prove our claim. Comparison of results obtained with the classical k-means and the present algorithm indicates usefulness of the new approach.

Design of Cellular Manufacturing System involves four major decisions: Cell formation (CF), Group layout (GL), Group scheduling (Gs) and Resource assignment (RA). These problems should be regarded, concurrently, in order to obtain an optimal solution in a CM environment. In this paper a two stage heuristic procedure is proposed for CF and RA decisions problem. The solution approach contains a heuristic multivariate clustering technique as the first stage to find the best machine-cluster center distances. Next in the second stage a new mathematical model based on extracted distances and also worker related issues, including salary, hiring, firing and cross-training is proposed. In order to verify and validate the performance of proposed approach a mathematical model considering the inter-intra cell part trips and also operator related issues are developed and some numerical examples are solved using Lingo Software. The analysis of results verifies the solution approach in both optimality and computational time aspects.