The need for an ensemble classifier is driven by better accuracy; reduced overfitting, increased robustness that copes with noisy data and reduced variance of individual models, combining the advantages and overcoming the drawbacks of the individual classifier. A comparison of different classifiers like Support Vector Machine (SVM), XGBoost, Random Forest (RF), Naive Bayes (NB), Convolutional Neural Network (CNN) and proposed Ensemble method used in the classification task was conducted. Among all the classifiers evaluated, CNN was found to be the most accurate having an accuracy rate of 93.7%. This indicates that CNN can identify complex data patterns that are also important for photo recognition and classification tasks. Nonetheless, NB and SVM only achieved medium results with accuracy rates of 82.66% and 85.6% respectively. These could have been due to either the complexity of data being handled or underlying assumptions made. RF and XGBoost demonstrated remarkable performances by employing ensemble learning methods as well as gradient-boosting approaches with accuracies of 83.33% and 90.7% respectively. The Ensemble method presented in this paper outperformed all individual models at an accuracy level of 95.5%, indicating that more than one technique is better when classifying correctly based on various resource allocations across techniques employed thereby improving such outcomes altogether by combining them. These results display the pros and cons of every classifier on the Plant Village dataset, giving vital data to improve plant disease classification and guide further research into precision farming and agricultural diagnostics.

Download article

• APA 7th style

• Chicago style

• IEEE style

• Vancouver style

Diabetes Mellitus (DM) is a persistent metabolic disorder which is characterized by increased blood glucose level in the blood stream. Initially, DM occurs while the insulin secretion in the pancreas has a disability to secrete or to use hormone for the metabolic process. Moreover, there are different types of DM depending on the physiological process, and the types include Type1 DM, Type2 DM and Gestational DM. Electrocardiography (ECG) waves are used to detect the abnormal heartbeats and cannot directly detect DM, but the wave abnormality can indicate the possibility and presence of DM. Whereas the Photoplethysmography (PPG) signals are a non-invasive method used to detect changes in  blood volume that can monitor BG changes. Furthermore, the detection and classification of DM using PPG and ECG can involve analyzing the functional performance of these modalities. By extracting the features like R wave (W1) and QRS complex (W2) in the ECG signals and Pulse Width (S1) and Pulse Amplitude Variation (S2) can detect DM and can be classified into DM and Non-DM. The authors propose a Novel architecture in the basis of Encoder Decoder structure named as Obstructive Encoder Decoder module. This module extracts the specific features and the proposed novel Obstructive Erasing Module remove the remaining artifacts and then the extracted features are fed into the Multi-Uni-Net for the fusion of the two modalities and the fused image is classified using EXplainable Machine Learning (EX-ML). From this classification the performance metrics like Accuracy, Precision, Recall, F1-Score and AUC can be determined.

Download article

• APA 7th style

• Chicago style

• IEEE style

• Vancouver style

This paper presents a comprehensive computational system designed to evaluate the undetectability of video steganography from human perspective. The system assesses the perceptibility of steganographic modifications to the human eye while simultaneously determining the minimum encoding level required for successful automated decoding of hidden messages. The proposed architecture comprises four subsystems: steganogram database preparation, human evaluation, automated decoding, and comparative analysis. The system was tested using example steganographic techniques applied to a dataset of video files. Experimental results revealed the thresholds of human-level undetectability and automated decoding for each technique, enabling the identification of critical differences between human and algorithmic detection capabilities. This research contributes to the field of steganography by offering a novel framework for evaluating the trade-offs between human perception and automated decoding in video-based information hiding. The system serves as a tool for advancing the development of more secure and reliable video steganographic techniques.

• APA 7th style

• Chicago style

• IEEE style

• Vancouver style

According to the World Health Organization, the Global Mental Health Report estimated that between 251 and 310 million individuals worldwide experienced depression during the first year of the COVID-19 pandemic. Most methods for detecting depression rely on clinical diagnoses and surveys. However, the American Psychiatric Association reports that over 50% of patients do not receive appropriate treatment. This study aims to utilize machine learning and pupil diameter features to identify depression and evaluate the accuracy of these classifiers in comparison to our previous deep learning model. While limited research has explored the use of pupillary diameter as a classification tool for distinguishing between individuals with and without depression, several studies have focused on EEG signals for this purpose. The study involved 58 participants, with 29 classified as depressed and 29 as healthy. The classification was based on statistical features extracted from the Hilbert-Huang Transform. Results showed a significant improvement in average accuracy compared to the authors’ prior work, with the current study achieving 77.72% accuracy, compared to 64.78% in their previous research. Machine learning methods, particularly Bagging, outperformed deep learning models such as AlexNet when classifying data from the left and right eyes individually (90.91% vs. 78.57% for the left eye; 90.91% vs. 71.43% for the right eye). However, when combining data from both eyes, deep learning using AlexNet demonstrated superior performance (98.28% accuracy compared to 93.75% using Bagging with statistical features from both eyes). Despite the higher accuracy of deep learning, machine learning is recommended for its faster execution times.

Download article

• APA 7th style

• Chicago style

• IEEE style

• Vancouver style

This study aims to map three main benthic habitats (coral, seagrass, and sand) in Kapota Atoll (Wakatobi, Indonesia) using a single-beam echosounder (SBES) Simrad EK15. The acoustic data were processed using Sonar5-Pro software. Eight acoustic parameters were used as input for the classification and prediction of benthic habitats, including depth (D), five acoustic parameters of the first echo (BD, BP, AttSv1, DecSv1, and AttDecSv1), and cumulative energy of the second and third echoes (AttDecSv2 and AttDecSv3). The classification and prediction process of benthic habitats uses two machine learning algorithms, Random Forest (RF) and Support Vector Machine (SVM), in XLSTAT Basic+ software. The study results show that 49 combinations of acoustic parameters produce benthic habitat maps that meet the minimum accuracy standards for benthic habitat mapping (≥60%). Using eight acoustic parameters produces a more accurate benthic habitat map than using only two main SBES parameters (DecSv1 and AttDecSv2 parameters or E1 and E2 in the RoxAnn system indicating the roughness and hardness indices). The RF and SVM algorithms produce benthic habitat maps with the highest accuracy of 79.33% and 78.67%, respectively. Each acoustic parameter has a different importance for the classification of benthic habitats, where the order of importance of each acoustic parameter in the overall classification follows the following order: AttDecSv2 > D > DecSv1 > BD > AttDecSv3 > AttSv1 > AttDecSv1 > BP. Overall, using more acoustic parameters can significantly improve the accuracy of benthic habitat maps.

• APA 7th style

• Chicago style

• IEEE style

• Vancouver style

This research investigates the enhancement of Artificial Immune Systems (AIS) for solving the Traveling Salesman Problem (TSP) through hybridization with Neighborhood Improvement (NI) and parameter fine-tuning. Two main experiments were conducted: Experiment A identified the optimal integration points for NI within AIS, revealing that position 2 (AIS+NIpos2) improved solution quality by an average of 27.78% compared to other positions. Experiment B benchmarked AIS performance with various enhancement techniques. Using symmetric and asymmetric TSP datasets, the results showed that integrating NI at strategic points and fine-tuning parameters boosted AIS performance by up to 46.27% in some cases. The hybrid and fine-tuned version of AIS (AIS-th) consistently provided the best solution quality, with up to a 50.36% improvement, though it required more computational time. These findings emphasize the importance of strategic combinations and fine-tuning for creating effective optimization algorithms.

• APA 7th style

• Chicago style

• IEEE style

• Vancouver style

This study investigates the application of large language models, particularly ChatGPT, in the extraction and structuring of medical information from free-text patient reports. The authors explore two distinct methods: a zero-shot extraction approach and a schema-based extraction approach. The dataset, consisting of 1230 anonymized French medical reports from the Department of Neonatology of the Mohammed VI University Hospital, served as the basis for these experiments. The findings indicate that while ChatGPT demonstrates a significant capability in structuring medical data, certain challenges remain, particularly with complex and non-standardized text formats. The authors evaluate the model's performance using precision, recall, and F1 score metrics, providing a comprehensive assessment of its applicability in clinical settings.

• APA 7th style

• Chicago style

• IEEE style

• Vancouver style

A distributed type of database where digital data is stored as blocks chained together is called a Blockchain. Each block consists of several transactions, authenticated using cryptographic keys, and approved by a group of validators. Hundreds of different Blockchain solutions have been proposed over the years, proving that they attract research and business interest. In this article, the authors present a generic vocabulary for unifying different terminologies used by various researchers in the field, followed by a review and evaluation of several research works presenting Blockchain-based solutions. A set of criteria regarding usefulness and suitability of adopting a Blockchain application are distinguished in these works. A method to examine their applicability is also discussed. Cryptocurrencies and supply chains, the two most well-known Blockchain uses, are considered and examined to assess how important these criteria are in these two use cases.

• APA 7th style

• Chicago style

• IEEE style

• Vancouver style

This paper evaluates the feasibility of deploying locally-run Large Language Models (LLMs) for retrieval-augmented question answering (RAG-QA) over internal knowledge bases in small and medium enterprises (SMEs), with a focus on Polish-language datasets. The study benchmarks eight popular open-source and source-available LLMs, including Google’s Gemma-9B and Speakleash’s Bielik-11B, assessing their performance across closed, open, and detailed question types, with metrics for language quality, factual accuracy, response stability, and processing efficiency. The results highlight that desktop-class LLMs, though limited in factual accuracy (with top scores of 45% and 43% for Gemma and Bielik, respectively), hold promise for early-stage enterprise implementations. Key findings include Bielik's superior performance on open-ended and detailed questions and Gemma's efficiency and reliability in closed-type queries. Distribution analyses revealed variability in model outputs, with Bielik and Gemma showing the most stable response distributions. This research underscores the potential of offline-capable LLMs as cost-effective tools for secure knowledge management in Polish SMEs.

• APA 7th style

• Chicago style

• IEEE style

• Vancouver style

Magnetic resonance imaging (MRI) is a key method for imaging human tissues and organs. The accuracy of medical diagnosis is greatly affected by the quality of MRI images. Sometimes, MRI images are obtained blurry due to various inevitable constraints related to the imaging equipment, which affects the detection of important features in the image. Several sharpening methods were introduced, but not all were successful in this task, as artifacts may be introduced, contrast may be changed, and high complexity may be involved. Thus, this paper introduces a guided-subsumed unsharp mask filter (GSUM) to improve the sharpness of MRI images. The GSUM utilizes an improved guided filter instead of the low-pass Gaussian filter and a dynamic sharpening parameter. The improved guided filter employs a hybrid procedure instead of the mean filter in the smoothing process and relies on an adaptive regularization parameter. The applied modifications eliminated the overshooting and halo effects of the original unsharp masking and the guided filter, resulting in better-quality images. The GSUM was tested with real-blurry MRI images, evaluated using three no-reference metrics, and compared with six other algorithms. The metric scores indicate that the proposed filter can surpass existing methods, as it produced better results with average readings of 24.2074 in PIQE, 0.6878 in BLUR, and 5.7944 in FISH. It also scored a fast computation time, averaging 0.3384 seconds.

• APA 7th style

• Chicago style

• IEEE style

• Vancouver style

One of the most important goals in image segmentation is the process of separating the object parts from the image background. Image segmentation is also a fundamental stage in the development of other image applications such as object recognition, target tracking, computer vision, and biomedical image processing. Interactive image segmentation methods with additional user interaction are still popular in research. Interactive image segmentation aims to provide additional information through simple interactions, especially in images with complex objects. Interactive image segmentation with region merging processes has drawbacks, one of which is suboptimal region splitting due to soft color shades, blurred contours, and uneven lighting, referred to in this study as ambiguous regions. However, in the fuzzy region initialization stage after obtaining values from the marker process, there is a possibility of missing or suboptimal determination of fuzzy regions. This is because it only takes the highest gray level value for the background marker and the lowest gray level value for the object marker. In this study, fuzzy region merging using hierarchical clustering is proposed to find optimal initialization for fuzzy regions in image segmentation. Based on the experimental results, the proposed method can achieve optimal segmentation with an average misclassification error value of 2.62% for Natural Images and 9.33% for Dental Images.

• APA 7th style

• Chicago style

• IEEE style

• Vancouver style