Open Access

Article

Article ID: 4241

Image-Based Soil Nutrient Prediction Using Machine Learning and Deep Learning Models

by Preity Mishra, S.K. Chaulya, Vikash Kumar, Naresh Kumar, Abhinek Agarwal, Anant Stephen Besra

Computing and Artificial Intelligence, Vol.4, No.3, 2026;

Recent advances in precision agriculture have enabled the use of machine learning and image processing for rapid soil nutrient assessment. This study presents a non-destructive approach for estimating soil nutrient content from digital soil images captured using a camera and analyzed through specialized software. A dataset of 1,500 soil images was processed using color correction, noise reduction, and feature extraction techniques to improve data quality. Five machine learning models, Random Forest, Extreme Gradient Boosting (XGBoost), Light Gradient Boosting Machine (LightGBM), Gradient Boosting Machine (GBM), and a Stacking Regressor, were evaluated for predicting nitrogen, phosphorus, and potassium levels. A Convolutional Neural Network (CNN) based deep learning model was served as a benchmark. No single model was best for every nutrient and metric. For nitrogen, GBM achieved the lowest test MAE (0.1652), whereas the Stacking Regressor achieved the highest test R² (0.9886). For potassium, Random Forest achieved the lowest test MAE (0.0051), whereas XGBoost achieved the highest test R² (0.9999). For phosphorus, XGBoost achieved both the lowest test MAE (0.1558) and the highest test R² (0.9929). Thus, the best-performing model depends on the nutrient and evaluation metric. Further validation using independent field datasets is required to establish robustness under diverse environmental conditions.

Open Access

Article

Article ID: 4266

Enhancing the validity of probability prompts using four generative artificial intelligence tools

by Ayşe Fethiye Gemci, Amanda Gantt Sawyer, Haley Smiley, Richard Busi, Juliana Galano

Computing and Artificial Intelligence, Vol.4, No.3, 2026;

Generative Artificial Intelligence (genAI) has become a large influence on students and teachers in the mathematics classroom. Specifically, researchers found that genAI tools could solve probability questions such as “What is the probability of drawing the Queen of Hearts from a standard deck of playing cards on the first try?” These researchers also stated that the genAI tools could not be trusted when solving, “Three cards are lying face down in front of me. Two are face cards, and there are no spades. No card is next to a card of the same color. What are the odds that the card on the left is the Queen of Hearts?” However, it is unclear whether that statement is true. Thus, we verified these findings by investigating four genAI tools (ChatGPT, Gemini, Co-Pilot, and Claude AI) on their ability to complete the identified probability task to determine when genAI can be trusted to perform probability calculation. Using the results of this investigation, we examined these genAI models using their responses and the predictions made by the genAI trainer. The investigation implies that developing mathematical validity requires intentional instruction, including teaching students how to compare genAI’s explanations, conducting GenAI Triangulation Process, cross-checking results using independent methods or alternative tools, and justifying whether an answer is mathematically sound.

Open Access

Article

Article ID: 4251

A generalized instance weighting principal component analysis framework

by Ernest Domanaanmwi Ganaa

Computing and Artificial Intelligence, Vol.4, No.2, 2026;

As a classical algorithm, principal component analysis (PCA) considers all data points as equally important in pursuing projections, although some data points may be more useful than others. This consequently makes traditional PCA sensitive to noisy or corrupt data. Another challenge faced by PCA is the problem of dealing with missing observations. These two challenges may significantly affect its performance negatively. In this study, we present a generalized instance weighting PCA (IWPCA) algorithm that innovatively addresses the problems of the traditional PCA outlined above. Instead of seeking orthogonal projections directly in PCA, data is embedded in the projection space to remove a random scaling factor in the projection space and therefore, solves the issue of the missing observations challenge. To achieve better projections, a sample instance is further introduced into the model. Two approaches are then used to weigh instances from geometric and data learning views. Thus, minimizing the impact of outliers and noisy instances in the proposed algorithm. This makes the proposed model a special form of the classical PCA where instances are weighed in pursuing projections, thus improving performance by enhancing projections. Extensive experimental evaluations indicate the proposed method has superior performance over all the comparative methods.

Open Access

Article

Article ID: 4398

YOLOv26-based automated crater detection and geometric measurement using Chandrayaan-2 Orbiter High Resolution Camera (OHRC) imagery

by Meenakshi Sarkar, Prit Gajjar

Computing and Artificial Intelligence, Vol.4, No.2, 2026;

Accurate detection and measurement of lunar craters are crucial for a better understanding of the geological history and evolution of the Moon’s surface. Manually detecting and measuring craters using high-resolution images is not only tedious but may also involve a certain amount of subjectivity. What makes this proposal innovative is the combination of the You Only Look Once (YOLO) algorithm with the Segment Anything Model for accurate segmentation of craters’ pixels and measurement of their geometrical properties. In this framework, the locations of the craters are automatically detected using the YOLOv26 algorithm, followed by accurate determination of the crater regions using the Segment Anything Model (SAM). Finally, the accurate determination of the diameter, radius, and surface area of the craters is carried out using the pixel measurement and scale factor conversion technique. This approach will be trained and tested using a dataset comprised of about 12,000 images obtained by dividing the Orbiter High Resolution Camera (OHRC) lunar imagery into image patches of 640 × 640 dimensions and divided in the ratio of training set: validation set: test set as 70:20:10. This model gave a precision value of 91.5%, a recall of 88.6%, mAP@50 of 95.6%, and mAP@50–95 of 88.6%, thereby outperforming all other versions of YOLO that were analysed and showing strong competency.

Open Access

Article

Article ID: 4439

Static vs. dynamic cloud scheduling: A benchmark driven performance and energy study

by Saba Naz, Altaf Hussain

Computing and Artificial Intelligence, Vol.4, No.2, 2026;

Major multinational corporations, including Amazon, Microsoft, IBM, and Google, continue to advance network and computational infrastructures through the deployment of highly efficient data centers worldwide to strengthen their cloud computing services. Despite these technological advancements, cloud computing environments still face several critical challenges, such as optimal resource utilization, cost efficiency, security, fault tolerance, scalability, and energy consumption. For organizations operating private clouds or delivering cloud-based services, the primary objectives include maximizing resource utilization, minimizing response and waiting times, increasing throughput and profitability, and enhancing the overall user experience, with energy efficiency being a particularly important concern. This study presents an empirical evaluation of various static and dynamic cloud job scheduling algorithms using the CloudSimPlus simulation framework. The algorithms are assessed using two well-established scientific benchmark datasets: HCSP and GoCJ instances. Experimental results reveal that the Resource-Aware Load Balancing Algorithm (RALBA) emerges as the most energy-efficient static scheduling approach, achieving reduced makespan, improved resource utilization, and lower energy consumption. In contrast, the Max–Min algorithm exhibits the highest makespan, the lowest throughput, inefficient resource utilization, and the greatest energy consumption. Among the dynamic scheduling techniques, DE-RALBA demonstrates superior resource utilization and competitive makespan performance while maintaining relatively efficient energy usage. However, the results also indicate that dynamic scheduling approaches may incur higher energy consumption depending on workload characteristics and rescheduling overhead. Unlike prior studies that evaluate scheduling strategies in isolation, this work provides a unified, benchmark-driven, and energy-aware comparison of static and dynamic scheduling paradigms, enabling deeper insights into performance–energy trade-offs.

Open Access

Article

Article ID: 4448

Epigenetic regulation for dynamic UAV-swarm optimization—EpiSwarm

by Jordi Vallverdú

Computing and Artificial Intelligence, Vol.4, No.1, 2026;

Evolutionary algorithms adapt mainly through selection, recombination, and mutation, which in changing environments forces costly re-optimization and slow recovery after disruption. We present EpiSwarm, an evolutionary algorithm for dynamic industrial UAV-swarm inspection in which each candidate solution carries a persistent genotype and a fast-changing, partially heritable epigenotype. The genotype encodes mission structure; the epigenotype is a decaying memory of recent environmental stress that modulates task activation, route repair, operator intensity, and energy-aware reassignment. On a UAV-swarm inspection benchmark (8 UAVs, 45 tasks, three volatility regimes, n = 10 replications) EpiSwarm reduces mean replanning churn by about 60% (from 12–15 to 5–6 changes per epoch) and improves mission utility by 10.7–14.2% over the best of four baselines—including a memory-archive GA—with this advantage robust across four objective-weight specifications (p between 0.009 and 0.052). A matched paired ablation (n = 20) shows that the benefit of transgenerational inheritance is regime-dependent: it is significant at low volatility (p = 0.025) and vanishes as disruptions accelerate. We show that this crossover is not incidental but follows a closed-form memory-horizon condition, τepi ≲ τenv/g, that predicts a priori the regime in which inherited memory ceases to be informative—a computational instance of the established evolutionary-biology result that the value of transgenerational information is set by environmental autocorrelation. The stress-driven update is directionally positive but not statistically separable at this sample size, and is reported as such. EpiSwarm therefore contributes a quantitatively predictable, low-churn regulatory mechanism for dynamic swarm optimization rather than a universally beneficial add-on.