AI, Volume 6, Issue 2 (February 2025) – 7 articles

Predicting crop performance is key to decision making for farmers and business owners. Tacna is the main olive-producing region in Perú, with an annual yield of 6.4 t/ha, mainly of the Sevillana variety. Recently, olive production levels have fluctuated due to severe weather conditions and disease outbreaks. These climatic phenomena are expected to continue in the coming years. The objective of the study was to evaluate the performance of the model in natural and specific environments of the olive grove and counting olive fruits using CNNs from images. Among the models evaluated, YOLOv8m proved to be the most effective (94.960), followed by YOLOv8s, Faster R-CNN and RetinaNet. For the mAP50-95 metric, YOLOv8m was also the most effective (0.775). YOLOv8m achieved the best performance with an RMSE of 402.458 and a coefficient of determination R² of (0.944), indicating a high correlation with the actual fruit count. As part of this study, a novel olive fruit dataset was developed to capture the variability under different fruit conditions. Concluded that the predicting crop from images requires consideration of field imaging conditions, color tones, and the similarity between olives and leaves. Full article

This study explores the extraction of remote Photoplethysmography (rPPG) signals from images using various neural network architectures, addressing the challenge of accurate signal estimation in biomedical contexts. The objective is to evaluate the effectiveness of different models in capturing rPPG signals from dataset snapshots. Two training strategies were investigated: pre-training models with only the fully connected layer being fine-tuned and training the entire network from scratch. The analysis reveals that models trained from scratch consistently outperform their pre-trained counterparts in extracting rPPG signals. Among the architectures assessed, DenseNet121 demonstrated superior performance, offering the most reliable results in this context. These findings underscore the potential of neural networks in advancing rPPG signal extraction, which has promising applications in fields such as clinical monitoring and personalized medical care. This study contributes to the integration of advanced imaging techniques and neural network-based analysis in biomedical engineering, paving the way for more robust and efficient methodologies. Full article

The combinatorial fusion cascade provides a surprisingly simple and complete explanation for the origin of the genetic code based on competing protocodes. Although its molecular basis is only beginning to be uncovered, it represents a natural pattern of information generation from initial signals and has potential applications in designing more-efficient neural networks. By utilizing the properties of the combinatorial fusion cascade, we demonstrate its embedding into deep neural networks with sequential fully connected layers using the dynamic matrix method and compare the resulting modifications. We observe that the Fiedler Laplacian eigenvector of a combinatorial cascade neural network does not reflect the cascade architecture. Instead, eigenvectors associated with the cascade structure exhibit higher Laplacian eigenvalues and are distributed widely across the network. We analyze a text classification model consisting of two sequential transformer layers with an embedded cascade architecture. The cascade shows a significant influence on the classifier’s performance, particularly when trained on a reduced dataset (approximately 3% of the original). The properties of the combinatorial fusion cascade are further examined for their application in training neural networks without relying on traditional error backpropagation. Full article

This study advances AI-powered news delivery by introducing an innovative chatbot capable of providing personalized news summaries and real-time event analysis. This approach addressed a critical gap identified through a comprehensive review of 52 AI chatbot studies. Unlike prior models limited to static information retrieval or predefined interactions, this chatbot harnesses generative AI and real-time data integration to deliver a dynamic and tailored news experience. Its unique architecture combines conversational AI, robotic process automation (RPA), a comprehensive news database (989,432 reports from 2342 sources spanning 27 October 2023 to 30 September 2024), and a large language model (LLM). Within this architecture, LLM generates dynamic queries against the News database for obtain tailored News for the users. Hence, this approach interprets user intent, and delivers LLM-based summaries of the fetched tailored news. Empirical testing with 35 users across 321 diverse news queries validated its robustness in navigating a combinatorial classification space of 53,916,650 potential news categorizations, achieving an F1-score of 0.97, recall of 0.99, and precision of 0.96. Deployed on Microsoft Teams and as a standalone web app, this research lays the foundation for transformative AI applications in news analysis, promising to revolutionize news consumption and empower a more informed citizenry. Full article

Accuratesnake species identification is essential for effective snakebite management, particularly in regions like Morocco, where approximately 400 snakebite incidents are reported annually, with a case fatality rate of 7.2%. Identifying venomous snakes promptly can significantly improve treatment outcomes by enabling the timely administration of specific antivenoms. However, the absence of comprehensive databases and rapid identification tools for Moroccan snake species poses challenges to effective clinical responses. This study presents a deep learning-based approach for the automated identification of Moroccan snake species. Several architectures, including VGG-19, VGG-16, and EfficientNet B0, were evaluated for their classification performance. EfficientNet B0 emerged as the most effective model, achieving an accuracy of 92.23% and an F1-score of 93.67%. After training on the SnakeCLEF 2021 dataset and fine-tuning with a specialized local dataset, the model attained a validation accuracy of 94% and an F1-score of 95.86%. To ensure practical applicability, the final model was deployed on a web platform, enabling the rapid and accurate identification of snake species via image uploads. This platform serves as a valuable tool for healthcare professionals and the general public, facilitating improved clinical response and educational awareness. This study highlights the potential of AI-driven solutions to address challenges in snakebite identification and management, offering a scalable approach for regions with limited resources and high snakebite prevalence. Full article

The .NET framework is widely used for software development, making it a target for a significant number of malware attacks by developing malicious executables. Previous studies on malware detection often relied on developing generic detection methods for Windows malware that were not tailored to the unique characteristics of .NET executables. As a result, there remains a significant knowledge gap regarding the development of effective detection methods tailored to .NET malware. This work introduces a novel framework for detecting malicious .NET executables using statically extracted method names. To address the lack of datasets focused exclusively on .NET malware, a new dataset consisting of both malicious and benign .NET executable features was created. Our approach involves decompiling .NET executables, parsing the resulting code, and extracting standard .NET method names. Subsequently, feature selection techniques were applied to filter out less relevant method names. The performance of six machine learning models—XGBoost, random forest, K-nearest neighbor (KNN), support vector machine (SVM), logistic regression, and naïve Bayes—was compared. The results indicate that XGBoost outperforms the other models, achieving an accuracy of 96.16% and an F1-score of 96.15%. The experimental results show that standard .NET method names are reliable features for detecting .NET malware. Full article

This study introduces a novel evaluation framework for predicting web page performance, utilizing state-of-the-art machine learning algorithms to enhance the accuracy and efficiency of web quality assessment. We systematically identify and analyze 59 key attributes that influence website performance, derived from an extensive literature review spanning from 2010 to 2024. By integrating a comprehensive set of performance metrics—encompassing usability, accessibility, content relevance, visual appeal, and technical performance—our framework transcends traditional methods that often rely on limited indicators. Employing various classification algorithms, including Support Vector Machines (SVMs), Logistic Regression, and Random Forest, we compare their effectiveness on both original and feature-selected datasets. Our findings reveal that SVMs achieved the highest predictive accuracy of 89% with feature selection, compared to 87% without feature selection. Similarly, Random Forest models showed a slight improvement, reaching 81% with feature selection versus 80% without. The application of feature selection techniques significantly enhances model performance, demonstrating the importance of focusing on impactful predictors. This research addresses critical gaps in the existing literature by proposing a methodology that utilizes newly extracted features, making it adaptable for evaluating the performance of various website types. The integration of automated tools for evaluation and predictive capabilities allows for proactive identification of potential performance issues, facilitating informed decision-making during the design and development phases. By bridging the gap between predictive modeling and optimization, this study contributes valuable insights to practitioners and researchers alike, establishing new benchmarks for future investigations in web page performance evaluation. Full article