Comparison of machine learning models in forecasting different ENSO types

Physica Scripta, Journal Year: 2024, Volume and Issue: 99(8), P. 086007 - 086007

Published: July 20, 2024

Abstract Accurate forecasting of the El Niño Southern Oscillation (ENSO) plays a critical role in mitigating impacts extreme weather conditions linked to ENSO variability on ecosystems. This study evaluates performance six machine learning models two types: Central Pacific (Niño 4 index) and East 3.4 index). The analyzed include Feed Forward Neural Network (FFNN), Long Short-term Memory (LSTM) neural network, eXtreme Gradient Boosting Regressor, K-Nearest Neighbors Support Vector using index lagged by months as predictor. were trained monthly indices from 1870 1992 tested 1993 2023. We also assess relative predictability types. Events defined when exceeded ± 0.4 . Our evaluation during testing period reveals that for models, deep network (LSTM FFNN) demonstrated superior at 6-month lead time. Furthermore, all achieved impressive all-season correlations ranging 0.93 0.97 threat score phases between 0.71 0.88 events, 0.72 events. types depended model strength event. Considering both phases, La Niña events forecasted with higher accuracy besides notably fell short capturing 2015/2016 These results highlight potential particularly approaches, skillful forecasting, leveraging its historical data.

Language: Английский

---

Comparison of Explainable Artificial Intelligence Model and Radiologist Review Performances to Detect Breast Cancer in 752 Patients

Journal of Ultrasound in Medicine, Journal Year: 2024, Volume and Issue: 43(11), P. 2051 - 2068

Published: July 25, 2024

Objectives Breast cancer is a type of caused by the uncontrolled growth cells in breast tissue. In few cases, erroneous diagnosis specialists and unnecessary biopsies can lead to various negative consequences. some radiologic examinations or clinical findings may raise suspicion cancer, but subsequent detailed evaluations not confirm cancer. addition causing anxiety stress patients, such also biopsy procedures, which are painful, expensive, prone misdiagnosis. Therefore, there need for development more accurate reliable methods diagnosis. Methods this study, we proposed an artificial intelligence (AI)‐based method automatically classifying solid mass lesions as benign vs malignant. new dataset (Breast‐XD) was created with 791 belonging 752 different patients aged 18 85 years, were examined experienced radiologists between 2017 2022. Results Six classifiers, support vector machine (SVM), K‐nearest neighbor (K‐NN), random forest (RF), decision tree (DT), logistic regression (LR), XGBoost, trained on training samples Breast‐XD dataset. Then, each classifier made predictions 159 test data that it had seen before. The highest classification result obtained using explainable XGBoost model (X 2 GAI) accuracy 94.34%. An structure implemented build reliability developed model. Conclusions results X GAI compared according from biopsy. It observed our performed well cases where gave false positive results.

Language: Английский

---

Development and application of machine learning models in US consumer price index forecasting: Analysis of a hybrid approach

Data Science in Finance and Economics, Journal Year: 2024, Volume and Issue: 4(4), P. 469 - 513

Published: Jan. 1, 2024

<p>This study aims to apply advanced machine-learning models and hybrid approaches improve the forecasting accuracy of US Consumer Price Index (CPI). The examined performance LSTM, MARS, XGBoost, LSTM-MARS, LSTM-XGBoost using a large time-series data from January 1974 October 2023. were combined with key economic indicators US, hyperparameters optimized genetic algorithm Bayesian optimization methods. According VAR model results, variables such as past values CPI, oil prices (OP), gross domestic product (GDP) have strong significant effects on CPI. In particular, provided superior in CPI forecasts compared other was found perform best by establishing relationships federal funds rate (FFER) GDP. These results suggest that can significantly provide valuable insights for policymakers, investors, market analysts.</p>

Language: Английский

---

Explainable Artificial Intelligence to Predict the Water Status of Cotton (Gossypium hirsutum L., 1763) from Sentinel-2 Images in the Mediterranean Area

Plants, Journal Year: 2024, Volume and Issue: 13(23), P. 3325 - 3325

Published: Nov. 27, 2024

Climate change and water scarcity bring significant challenges to agricultural systems in the Mediterranean region. Novel methods are required rapidly monitor stress of crop avoid qualitative losses products. This study aimed predict stem potential cotton (Gossypium hirsutum L., 1763) using Sentinel-2 satellite imagery machine learning techniques enhance monitoring management cotton’s status. The research was conducted Rutigliano, Southern Italy, during 2023 growing season. Different algorithms, including random forest, support vector regression, extreme gradient boosting, were evaluated spectral bands as predictors. models’ performance assessed R2 root mean square error (RMSE). Feature importance analyzed permutation SHAP methods. forest model bands’ reflectance predictors showed highest performance, with an 0.75 (±0.07) RMSE 0.11 (±0.02). XGBoost (R2: 0.73 ± 0.09, RMSE: 0.12 0.02) AdaBoost 0.67 0.08, 0.13 followed performance. Visible (blue red) red edge identified most influential trained RF used seasonal trend potential, detecting periods acute moderate stress. approach demonstrates prospective for high-frequency, non-invasive status, which could smart irrigation strategies improve use efficiency production.

Language: Английский

---

Machine Learning-Driven Transcriptome Analysis of Keratoconus for Predictive Biomarker Identification

Biomedicines, Journal Year: 2025, Volume and Issue: 13(5), P. 1032 - 1032

Published: April 24, 2025

Background: Keratoconus (KTCN) is a multifactorial disease characterized by progressive corneal degeneration. Recent studies suggest that gene expression analysis of corneas may uncover potential novel biomarkers involved in matrix remodeling. However, identifying reliable combinations are linked to risk or progression remains significant challenge. Objective: This study employed multiple machine learning algorithms analyze the transcriptomes keratoconus patients, feature and their functional associations, with aim enhancing understanding pathogenesis. Methods: We analyzed GSE77938 (PRJNA312169) dataset for differential (DGE) performed set enrichment (GSEA) using Kyoto Encyclopedia Genes Genomes (KEGG) pathways identify enriched versus controls. Machine were then used sets, SHapley Additive exPlanations (SHAP) applied assess contribution key genes model’s predictions. Selected further through Gene Ontology (GO) explore roles biological processes cellular functions. Results: models, including XGBoost, Random Forest, Logistic Regression, SVM, identified important associated keratoconus, 15 notable appearing across such as IL1R1, JUN, CYBB, CXCR4, KRT13, KRT14, S100A8, S100A9, others. The under-expressed KTCN mechanical resistance epidermis (KRT14, KRT15) inflammation (S100A8/A9, CXCR4), compared GO highlighted S100A8/A9 complex its primarily related cytoskeleton organization, inflammation, immune response. Furthermore, we expanded our incorporating additional datasets from PRJNA636666 PRJNA1184491, thereby offering broader representation features increasing generalizability results diverse cohorts. Conclusions: differing sets XGBoost SVM reflect distinct but complementary aspects pathophysiology. Meanwhile, captured chemotactic regulators (e.g., suggesting upstream inflammatory signaling pathways. structural epithelial differentiation markers S100A8/A9), possibly reflecting downstream tissue remodeling stress responses. Our findings provide research platform evaluation learning-based approaches, valuable insights into pathogenesis therapeutic targets.

Language: Английский

---