This study presents a novel intelligent problem-solving method by strengthening Starfish Optimization Algorithm (SFOA) with roulette fitness-distance balance (RFDB) and quasi-opposition-based learning (QOBL). The SFOA-QOBL-RFDB is developed by making three effective modifications to the optimization scheme of the original SFOA. First, the QOBL strategy is integrated into the population initialization process. Secondly, the exploitation operator is redesigned using the RFDB method. Finally, an adaptive algorithmic parameter is employed for a soft transition between exploitation and exploration. The efficiency of the developed SFOA-QOBL-RFDB against the SFOA and 12 competing optimizers is validated on classical, CEC14, and CEC20 benchmark problems. The Wilcoxon test results show that SFOA-QOBL-RFDB demonstrates excellence in 21, 19 and 17 problems of the CEC14 benchmark suite for 30D, 50D and 100D search spaces, respectively. Based on the same test, the success rate of the developed optimizer in the CEC20 is 90
The extensive deployment of Internet of Things (IoT) devices has led to an increasing number of network security threats. Traditional intrusion detection systems often face challenges such as low detection accuracy and high false negative rates when dealing with complex, massive, and imbalanced IoT data. To address these issues, this paper proposes an intrusion detection model that integrates Deep Convolutional Neural Networks (DCNN) and the Focal Loss Function for IoT security detection, which aims to improve the recognition capability for minority-class attack behaviors. The model utilizes DCNN to detect stealthy and polymorphic intrusions in IoT networks, reducing the false positive rates. By applying Focal Loss, the model decreases the loss value while enhancing the training effectiveness, thus improving the detection performance for rare attack samples. SMOTE is employed to address the data imbalance problem, improving the model’s robustness and accuracy. To validate the model’s effectiveness, retraining and evaluation were conducted on several datasets, including NF-UNSW-NB15, NF-BoT-IoT, and NF-UNSW-NB15-v2. Experimental results show that the model outperforms existing mainstream models in terms of precision, recall, F1 score, and other metrics. In particular, the model achieves an accuracy of 99.67
In this paper, we present a novel one-shot federated clustering approach based on evolving fuzzy clustering and shuffler-based anonymization. In federated clustering, each round reveals additional model information to the server and can increase the opportunity to infer properties of local datasets, even when raw samples are never shared. Consequently, fewer client–server exchanges and unlinkability are important, so transmitted information cannot be attributed to any single client. Data-stream clustering algorithms, which learn in a single pass over the data, align well with the one-shot federated learning paradigm, where a single communication round can be sufficient. The proposed Evolving Federated Density Peak Clustering (eFDPC) method combines evolving fuzzy data-stream clustering and density peak clustering in a federated setting. Each client independently evolves micro-clusters from private data mini-batches through cluster addition and merging. To limit communication, density peak clustering is applied to the learned micro-clusters and only the selected peaks are transmitted, rather than the full local models. A shuffler-based layer is introduced between the clients and the server to anonymize contributions prior to reconstruction and global peak selection. We consider a peak-only variant, eFDPC- 𝒫 , as well as eFDPC- 𝒢 with server-side label propagation. Experiments on synthetic and real-world benchmark datasets show that eFDPC outperforms recent federated clustering baselines in one-shot settings and scales to higher-dimensional, non-IID data.
Human activity recognition (HAR) has become more important in recent times because of its applications in different fields like entertainment, security and surveillance, intelligent environments, and health. A considerable quantity of work has been achieved on HAR and investigators utilized dissimilar techniques, namely device-free, object-tagged, and wearable, to identify human actions. Nowadays, the deep learning model is extensively utilized for the progress of the HAR approach to identify the human activity patterns from the sensor data. This paper presents a Conditional Mutual Information-Guided Feature Reduction Approach for Human Activity Recognition Using Extreme Learning Machines (CMIFRA-HARELM) model. The CMIFRA-HARELM model aims to develop a real-time HAR system to assist disabled individuals by enhancing safety, and independent living through intelligent monitoring and support. Primarily, the data pre-processing stage implements the min-max scaling to transform input data into a suitable format. The dimensionality reduction method uses Conditional Mutual Information (CMI) technique, following an optimal subset of 26 informative features through eliminating the redundant attributes while preserving discriminative data. For the HAR classification process, the proposed CMIFRA-HARELM technique designs a multi-layer kernel meta-extreme learning machine system. At last, the moth-flame optimization algorithm fine-tunes the parameter values of ML-MetaKELM system and results in superior classification performance. The experimental assessment of the CMIFRA-HARELM system is verified on the benchmark dataset. The comprehensive results point out the significant solution of the CMIFRA-HARELM model to the HAR process with maximum accuracy of 99
In this paper, an evolving fuzzy clustering algorithm for time series is proposed, in which the clustering is performed based on computing the potential of the data, adopting the rate of variation and the density as variables within a multivariable Gaussian potential function criterion. The algorithm identifies clusters with time-varying prototypes using fuzzy covariance matrices and an exponential distance metric. Additionally, a sensitivity factor is employed to select centers of clusters from the potential of the data, a mutation-like mechanism is used for updating the clusters, and a crossover-like mechanism is used for merging the clusters. Computational results consider the implementation of the algorithm for evolving fuzzy clustering of data applied to the identification of SISO and MIMO Hammerstein dynamic systems, where the algorithm is employed in the data fuzzification process and in the creation of fuzzy rules in a fuzzy inference system that represents the static nonlinearity of the Hammerstein dynamic system. Experimental results consider the implementation of the algorithm for evolving fuzzy clustering of data applied to online identification of real-world dynamic systems (Thermal System and 2DOF Helicopter), where the algorithm is employed in the data fuzzification process and in the creation/updating of evolving fuzzy rules in an evolving fuzzy inference system. The identified fuzzy models present state-space submodels in the consequent propositions of the fuzzy rules, and their parametric identification is performed through OKID (Observer/Kalman Filter Identification), which computes the Markov parameters of the model from experimental data, and ERA (Eigensystem Realization Algorithm), which computes the matrices of the state-space submodels from the Markov parameters. In general, the results indicate that the proposed algorithm identifies clusters which are representative of the behavior of dynamic systems and obtains competitive performance compared to clustering algorithms in the literature.
