A detailed study of resampling algorithms for cyberattack classification in engineering applications.

The evolution of engineering applications is highly relevant in the context of protecting industrial systems. As industries are increasingly interconnected, the need for robust cybersecurity measures becomes paramount. Engineering informatics not only provides tools for knowledge representation and extraction but also affords a comprehensive spectrum of developing sophisticated cybersecurity solutions. However, safeguarding industrial systems poses a unique challenge due to the inherent heterogeneity of data within these environments. Together with this problem, it's crucial to acknowledge that datasets that simulate real cyberattacks within these diverse environments exhibit a high imbalance, often skewed towards certain types of traffics. This study proposes a system for addressing class imbalance in cybersecurity. To do this, three oversampling (SMOTE, Borderline1-SMOTE, and ADASYN) and five undersampling (random undersampling, cluster centroids, NearMiss, repeated edited nearest neighbor, and Tomek Links) methods are tested. Particularly, these balancing algorithms are used to generate one-vs-rest binary models and to develop a two-stage classification system. By doing so, this study aims to enhance the efficacy of cybersecurity measures ensuring a more comprehensive understanding and defense against the diverse range of threats encountered in industrial environments. Experimental results demonstrates the effectiveness of proposed system for cyberattack detection and classification among nine widely known cyberattacks.

MRI-computer vision on fresh and frozen-thawed beef: Optimization of methodology for classification and quality prediction.

This study aims to evaluate the capability of Magnetic Resonance Imaging (MRI) and computer vision techniques to classify fresh (raw F) (n = 12) and frozen-thawed (FT) (n = 12) beef and predict physico-chemical, texture and sensory characteristics by optimization the methodology for image analysis (algorithm) and data analysis (regressor), testing different algorithm-regressor combinations. The accuracy of the classification and prediction results especially depend on the algorithm. Different optimum combinations were found for classification (Fractal with CForest, RF or SVM) and prediction of quality parameters of raw FT (Fractal-CForest or Fractal-RF) and cooked FT samples (Classic-RF). Thus, the computational analysis of MRI, especially the algorithm to analyze the image, may be set as a function of the aim (classification or prediction) and of the type of sample (raw or cooked), while the analysed characteristic is not relevant. This study firstly showed the capability of MRI to classify beef (raw F vs. raw FT) and to determine quality characteristics in a non-destructive way.

Optimization of the image acquisition procedure in low-field MRI for non-destructive analysis of loin using predictive models.

The use of low-field magnetic resonance imaging (LF-MRI) scanners has increased in recent years. The low economic cost in comparison to high-field (HF-MRI) scanners and the ease of maintenance make this type of scanner the best choice for nonmedical purposes. However, LF-MRI scanners produce low-quality images, which encourages the identification of optimization procedures to generate the best possible images. In this paper, optimization of the image acquisition procedure for an LF-MRI scanner is presented, and predictive models are developed. The MRI acquisition procedure was optimized to determine the physicochemical characteristics of pork loin in a nondestructive way using MRI, feature extraction algorithms and data processing methods. The most critical parameters (relaxation times, repetition time, and echo time) of the LF-MRI scanner were optimized, presenting a procedure that could be easily reproduced in other environments or for other purposes. In addition, two feature extraction algorithms (gray level co-occurrence matrix (GLCM) and one point fractal texture algorithm (OPFTA)) were evaluated. The optimization procedure was validated by using several evaluation metrics, achieving reliable and accurate results (r > 0.85; weighted absolute percentage error (WAPE) lower than 0.1%; root mean square error of prediction (RMSEP) lower than 0.1%; true standard deviation (TSTD) lower than 2; and mean absolute error (MAE) lower than 2). These results support the high degree of feasibility and accuracy of the optimized procedure of LF-MRI acquisition. No other papers present a procedure to optimize the image acquisition process in LF-MRI. Eventually, the optimization procedure could be applied to other LF-MRI systems.