Integrating Remote Sensing and Soil Features for Enhanced Machine Learning-Based Corn Yield Prediction in the Southern US

Sensors, Journal Year: 2025, Volume and Issue: 25(2), P. 543 - 543

Published: Jan. 18, 2025

Efficient and reliable corn (Zea mays L.) yield prediction is important for varietal selection by plant breeders management decision-making growers. Unlike prior studies that focus mainly on county-level or controlled laboratory-scale areas, this study targets a production-scale area, better representing real-world agricultural conditions offering more practical relevance farmers. Therefore, the objective of our was to determine best combination vegetation indices abiotic factors predicting in rain-fed, identify most suitable growth stage estimation using machine learning, effective learning model estimation. Our used high-resolution (6 cm) aerial multispectral imagery. Sixty-two different predictors, including soil properties (sand, silt, clay percentages), slope, spectral bands (red, green, blue, red-edge, NIR), (GNDRE, NDRE, TGI), color-space indices, wavelengths were derived from data collected at seven (V4, V5, V6, V7, V9, V12, V14/VT) stages corn. Four regression algorithms evaluated prediction: linear regression, random forest, extreme gradient boosting, boosting regressor. A total 6865 values training 1716 validation. Results show that, forest method, V14/VT had predictions (RMSE 0.52 Mg/ha mean 10.19 Mg/ha), V6 still feasible. We concluded integrating factors, such as slope properties, significantly improved accuracy. Among TGI, HUE, GNDRE performed better. can help farmers crop consultants plan ahead future logistics through enhanced early-season support farm profitability sustainability.

Language: Английский

---

Advanced Brain Tumor Classification in MR Images Using Transfer Learning and Pre-Trained Deep CNN Models

Cancers, Journal Year: 2025, Volume and Issue: 17(1), P. 121 - 121

Published: Jan. 2, 2025

Background/Objectives: Brain tumor classification is a crucial task in medical diagnostics, as early and accurate detection can significantly improve patient outcomes. This study investigates the effectiveness of pre-trained deep learning models classifying brain MRI images into four categories: Glioma, Meningioma, Pituitary, No Tumor, aiming to enhance diagnostic process through automation. Methods: A publicly available Tumor dataset containing 7023 was used this research. The employs state-of-the-art models, including Xception, MobileNetV2, InceptionV3, ResNet50, VGG16, DenseNet121, which are fine-tuned using transfer learning, combination with advanced preprocessing data augmentation techniques. Transfer applied fine-tune optimize accuracy while minimizing computational requirements, ensuring efficiency real-world applications. Results: Among tested Xception emerged top performer, achieving weighted 98.73% F1 score 95.29%, demonstrating exceptional generalization capabilities. These proved particularly effective addressing class imbalances delivering consistent performance across various evaluation metrics, thus their suitability for clinical adoption. However, challenges persist improving recall Glioma Meningioma categories, black-box nature requires further attention interpretability trust settings. Conclusions: findings underscore transformative potential imaging, offering pathway toward more reliable, scalable, efficient tools. Future research will focus on expanding diversity, model explainability, validating settings support widespread adoption AI-driven systems healthcare ensure integration workflows.

Language: Английский

---

Assessment of environmental and socioeconomic drivers of urban stormwater microplastics using machine learning

Scientific Reports, Journal Year: 2025, Volume and Issue: 15(1)

Published: Feb. 21, 2025

Language: Английский

---

Recent Advances in Machine Learning for Building Envelopes: From Prediction to Optimization

Published: Jan. 1, 2025

Nowadays, advanced building envelopes not only need to meet traditional design requirements but also address emerging demands, such as achieving low-carbon transition of buildings and mitigating the urban heat island (UHI) effect. Given intricacy indoor conditions complexity variables, approaches can hardly keep pace with evolving demands. Therefore, integrating Artificial Intelligence (AI) into envelope is trending in recent years. This paper provides a holistic review research on machine learning (ML) design. Popular ML algorithms, data input requirements, output generation are first elucidated, aiming shed light selection appropriate algorithms for specific datasets achieve optimal outcomes. ML-involved studies related types (e.g., building-integrated photovoltaic (BIPV), green roofs, PCM-integrated walls, glazing systems, etc.) discussed. The further highlights capabilities AI technologies predicting parameters material properties, environmental impact) optimizing criteria minimizing energy consumption), from micro-scope (i.e., microenvironment) macro-scope impact heat). work anticipated yield valuable insights promoting AI-driven solutions tackle both conventional challenges sustainable development.

Language: Английский

---

Data Augmentation Strategies for Improved PM2.5 Forecasting Using Transformer Architectures

Atmosphere, Journal Year: 2025, Volume and Issue: 16(2), P. 127 - 127

Published: Jan. 24, 2025

Breathing in fine particulate matter of diameter less than 2.5 µm (PM2.5) greatly increases an individual’s risk cardiovascular and respiratory diseases. As climate change progresses, extreme weather events, including wildfires, are expected to increase, exacerbating air pollution. However, models often struggle capture pollution events due the rarity high PM2.5 levels training datasets. To address this, we implemented cluster-based undersampling trained Transformer improve event prediction using various cutoff thresholds (12.1 µg/m3 35.5 µg/m3) partial sampling ratios (10/90, 20/80, 30/70, 40/60, 50/50). Our results demonstrate that threshold, paired with a 20/80 ratio, achieved best performance, RMSE 2.080, MAE 1.386, R2 0.914, particularly excelling forecasting events. Overall, on augmented data significantly outperformed those original data, highlighting importance resampling techniques improving quality accuracy, especially for high-pollution scenarios. These findings provide critical insights into optimizing models, enabling more reliable predictions By advancing ability forecast levels, this study contributes development informed public health environmental policies mitigate impacts pollution, advanced technology building better digital twins.

Language: Английский

---

Machine learning-assisted evaluation of PVSOL software using a real-time rooftop PV system: a case study in Kocaeli, Turkey, with a focus on diffuse solar radiation

International Journal of Low-Carbon Technologies, Journal Year: 2025, Volume and Issue: 20, P. 223 - 233

Published: Jan. 1, 2025

Abstract Reducing energy-related CO2 emissions is vital for global climate targets, with Net Zero Energy Buildings (NZEBs) playing a key role. This study evaluates PVSOL software’s accuracy in simulating rooftop photovoltaic (PV) system an NZEB Kocaeli, Turkey. A machine learning model enhanced result reliability using local weather data. The system’s first-year performance ratio was 81.9%, close to the theoretical 84.53%. 435 600 USD investment expected be recovered 11.42 years, while predicts 14.9 years. findings confirm PVSOL’s PV systems, emphasizing their effectiveness reduction and energy transition efforts.

Language: Английский

---

Machine learning for predicting severe dengue in Puerto Rico

Infectious Diseases of Poverty, Journal Year: 2025, Volume and Issue: 14(1)

Published: Feb. 4, 2025

Distinguishing between non-severe and severe dengue is crucial for timely intervention reducing morbidity mortality. World Health Organization (WHO)-recommended warning signs offer a practical approach clinicians but have limited sensitivity specificity. This study aims to evaluate machine learning (ML) model performance compared WHO-recommended in predicting among laboratory-confirmed cases Puerto Rico. We analyzed data from Rico's Sentinel Enhanced Dengue Surveillance System (May 2012-August 2024), using 40 clinical, demographic, laboratory variables. Nine ML models, including Decision Trees, K-Nearest Neighbors, Naïve Bayes, Support Vector Machines, Artificial Neural Networks, AdaBoost, CatBoost, LightGBM, XGBoost, were trained fivefold cross-validation evaluated with area under the receiver operating characteristic curve (AUC-ROC), sensitivity, A subanalysis excluded hemoconcentration leukopenia assess resource-limited settings. An AUC-ROC value of 0.5 indicates no discriminative power, while values closer 1.0 reflect better performance. Among 1708 cases, 24.3% classified as severe. Gradient boosting algorithms achieved highest predictive performance, an 97.1% (95% CI: 96.0-98.3%) CatBoost full 40-variable feature set. Feature importance analysis identified (≥ 20% increase during illness or ≥ above baseline age sex), (white blood cell count < 4000/mm3), timing presentation at 4-6 days post-symptom onset key predictors. When excluding leukopenia, was 96.7% 95.5-98.0%), demonstrating minimal reduction Individual like abdominal pain restlessness had sensitivities 79.0% 64.6%, lower specificities 48.4% 59.1%, respectively. Combining 3 improved specificity (80.9%) maintaining moderate (78.6%), resulting 74.0%. especially gradient algorithms, outperformed traditional dengue. Integrating these models into clinical decision-support tools could help identify high-risk patients, guiding interventions hospitalization, monitoring, administration intravenous fluids. The confirmed models' applicability settings, where access may be limited.

Language: Английский

---

TransECA-Net: A Transformer-Based Model for Encrypted Traffic Classification

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(6), P. 2977 - 2977

Published: March 10, 2025

Encrypted network traffic classification remains a critical component in security monitoring. However, existing approaches face two fundamental limitations: (1) conventional methods rely on manual feature engineering and are inadequate handling high-dimensional features; (2) they lack the capability to capture dynamic temporal patterns. This paper introduces TransECA-Net, novel hybrid deep learning architecture that addresses these limitations through key innovations. First, we integrate ECA-Net modules with CNN enable automated extraction efficient dimension reduction via channel selection. Second, incorporate Transformer encoder model global dependencies multi-head self-attention, supplemented by residual connections for optimal gradient flow. Extensive experiments ISCX VPN-nonVPN dataset demonstrate superiority of our approach. TransECA-Net achieved an average accuracy 98.25% classifying 12 types encrypted traffic, outperforming classical baseline models such as 1D-CNN, + LSTM, TFE-GNN 6.2–14.8%. Additionally, it demonstrated 37.44–48.84% improvement convergence speed during training process. Our proposed framework presents new paradigm disentanglement representation learning. enables cybersecurity systems achieve fine-grained service identification (e.g., 98.9% VPN detection) real-time responsiveness (48.8% faster than methods), providing technical support combating emerging cybercrimes monitoring illegal transactions darknet networks contributing significantly adaptive systems.

Language: Английский

---

Enhancing network intrusion detection: a dual-ensemble approach with CTGAN-balanced data and weak classifiers

The Journal of Supercomputing, Journal Year: 2024, Volume and Issue: 80(11), P. 16301 - 16333

Published: April 10, 2024

Language: Английский

---

Deep learning framework with Bayesian data imputation for modelling and forecasting groundwater levels

Environmental Modelling & Software, Journal Year: 2024, Volume and Issue: 178, P. 106072 - 106072

Published: May 19, 2024

Language: Английский

---