Improvement and Research of CNN Hard Attention LSTM Image Classification Model Suitable for Water Quality Detection DOI
Lewei Lin,

Rudi Lei,

Weinan Dai

et al.

2021 18th International Computer Conference on Wavelet Active Media Technology and Information Processing (ICCWAMTIP), Journal Year: 2023, Volume and Issue: 23, P. 1 - 6

Published: Dec. 15, 2023

In order to improve the accuracy and efficiency of water quality detection, firstly, collected images samples are preprocessed data enhancement is carried out, pictures divided into five classes according color, so as construct a classification dataset; detection algorithm combining Hard Attention mechanism (Hard Attention), Convolutional Neural Networks (CNNs), Long Short-Term Memory (LSTMs) proposed. Because hard attention more interpretable than soft able focus on specific area in image, it better exclude complex background that sample image may contain, such leaves, stones, plants, etc. this experiment, was chosen handle task. addition, because color difference small, capture correlation between adjacent pixels prevent overfitting problems, CNN-LSTM model selected for improvement experiment. The paper mainly analyses way signals signals, tries explore performance Support Vector Machines (SVM), (CNN), (LSTM) their improved after introduction Mechanism experimental results show outperforms other three algorithms dataset analysis task based color.

Language: Английский

A Structural Credit Risk Model with Jumps Based on Uncertainty Theory DOI Creative Commons

Hong Huang,

Meng Jiang, Yufu Ning

et al.

Mathematics, Journal Year: 2025, Volume and Issue: 13(6), P. 897 - 897

Published: March 7, 2025

This study, within the framework of uncertainty theory, employs an uncertain differential equation with jumps to model asset value process a company, establishing structured credit risk that incorporates jumps. is applied pricing two types derivatives, yielding formulas for corporate zero-coupon bonds and Credit Default Swap (CDS). Through numerical analysis, we examine impact volatility jump magnitude on default uncertainty, as well influence CDS. The results indicate increase in levels significantly enhances expansion negative not only directly elevates but also leads significant price CDS through premium adjustment mechanism. Therefore, when assessing disturbance must be considered crucial factor.

Language: Английский

Citations

0
Leveraging BiLSTM-GAT for enhanced stock market prediction: a dual-graph approach to portfolio optimization DOI Creative Commons
Xiaojian Lu, Josiah Poon, Matloob Khushi

et al.

Applied Intelligence, Journal Year: 2025, Volume and Issue: 55(7)

Published: March 31, 2025

Language: Английский

Citations

0
A Variational-Mode-Decomposition-Cascaded Long Short-Term Memory with Attention Model for VIX Prediction DOI Creative Commons

Do-Hyeon Kim,

Dong-Jun Kim,

Sun‐Yong Choi

et al.

Applied Sciences, Journal Year: 2025, Volume and Issue: 15(10), P. 5630 - 5630

Published: May 18, 2025

Financial time-series forecasting presents a significant challenge due to the inherent volatility and complex patterns in market data. This study introduces novel framework that integrates Variational Mode Decomposition (VMD) with Cascaded Long Short-Term Memory (LSTM) network enhanced by an Attention mechanism. The primary objective is enhance predictive accuracy of VIX, key measure uncertainty, through advanced signal processing deep learning techniques. VMD employed as preprocessing step decompose financial data into multiple Intrinsic Functions (IMFs), effectively isolating short-term fluctuations from long-term trends. These decomposed features serve inputs LSTM model mechanism, which enables capture critical temporal dependencies, thereby improving performance. Experimental evaluations using VIX S&P 500 January 2020 December 2024 demonstrate superior capability proposed compared seven benchmark models. results highlight effectiveness combining decomposition techniques Attention-based architectures for forecasting. research contributes field introducing hybrid improves accuracy, enhances robustness against fluctuations, underscores importance mechanisms capturing essential dynamics.

Language: Английский

Citations

0
Credit Behavior Scorecards Based on TabTransformer and Model Interpretability Exploring DOI
Songyun Ye,

Yuwen Jiang,

Bo Chen

et al.

Published: Jan. 10, 2025

Language: Английский

Citations

0
Influencing the Variable Selection and Prediction of Carbon Emissions in China DOI Open Access

Zhiyong Chang,

Yunmeng Jiao,

Xiaojing Wang

et al.

Sustainability, Journal Year: 2023, Volume and Issue: 15(18), P. 13848 - 13848

Published: Sept. 18, 2023

In order to study the changing rule of carbon dioxide emissions in China, this paper systematically focused on their current situation, influencing factors, and future trends. Firstly, situations global China’s were presented via a visualization method characteristics analyzed; secondly, random forest regression model was used screen main factors affecting emissions. Considering different aspects emissions, 29 determined 6 according results model. Then, prediction for China established. The BP neural network model, multi-factor LSTM time series CNN-LSTM compared test set all them passed test. However, goodness fit about 0.01~0.02 higher than other two models MAE RMSE 0.01~0.03 lower those models. Thus, it selected predict predicted showed that peak will be around 2027 these between 12.9 billion tons 13.2 tons. Overall, puts forward reasonable suggestions low-carbon development provides reference an adjustment plan energy structure.

Language: Английский

Citations

6
Implementation of a Fusion Classification Model for Efficient Pen-Holding Posture Detection DOI Open Access
Xiaoping Wu, Yupeng Liu, Chu Zhang

et al.

Electronics, Journal Year: 2023, Volume and Issue: 12(10), P. 2208 - 2208

Published: May 12, 2023

Pen-holding postures (PHPs) can significantly affect the speed and quality of writing, incorrect lead to health problems. This paper presents experimentally implements a methodology for quickly recognizing correcting poor writing using digital dot matrix pen. The method first extracts basic handwriting information, including page number, coordinates, movement trajectory, pen tip pressure, stroke sequence, handling time. information is then used generate features that are fed into our proposed fusion classification model, which combines simple parameter-free attention module convolutional neural networks (CNNs) called NetworkSimAM, CNNs, an extension well-known long short-term memory (LTSM) Mogrifier LSTM or MLSTM. Finally, ends with step (Softmax) recognize type PHP. implemented achieves significant results through receiver operating characteristic (ROC) curves loss functions, recognition accuracy 72%, is, example, higher than single-stroke model (i.e., TabNet incorporating SimAM). obtained show promising solution provided accurate efficient PHP has potential improve while reducing problems induced by postures.

Language: Английский

Citations

5
The Data Analysis of Enterprise Operational Risk Prediction Under Machine Learning DOI Open Access
Yixin Zhao

Journal of Organizational and End User Computing, Journal Year: 2024, Volume and Issue: 36(1), P. 1 - 24

Published: Sept. 20, 2024

In the digital age, financial sector faces increasingly severe risk management challenges. Traditional methods often rely on historical data and statistical models, which struggle to cope with high volatility of market. These exhibit poor adaptability in rapidly changing markets fail meet demands terms accuracy reliability. To address these issues, this study proposes a law prediction model based deep learning—the WBIF model. This integrates Bidirectional Long Short-Term Memory (BiLSTM) Fully Convolutional Networks (FCN) employs Whale Optimization Algorithm (WOA) for parameter optimization. Experimental results show that compared traditional reduces Mean Absolute Error (MAE) by 51.73% UCI machine learning library dataset improves 12% Kaggle credit card fraud detection dataset.

Language: Английский

Citations

1
A Logistic Regression Based Credit Risk Assessment Using WoE Bining and Enhanced Feature Engineering Approach ANOVA and Chi-Square DOI
Vandana Sharma, Amit Singh, Ashendra Kumar Saxena

et al.

Published: Dec. 22, 2023

This research paper presents a comprehensive approach to build data-driven credit risk model using machine learning techniques. Despite advancements in assessment methods, loan defaults remain significant concern for financial institutions. work describes novel and robust architecture designed tackle modeling scorecard prediction the field. The outlines systematic an innovative framework encompassing data cleaning, feature engineering, evaluation various matrices. dataset is obtained from peer-to-peer lending platform (Lending Club) having more than 450,000 features. Feature selection conducted Chi-squared test ANOVA F-statistic. Subsequently, Weight of Evidence binning engineering are detailed optimize predictive power selected trained logistic regression with class weight balancing. Following this, developed based on coefficients Loan approval cut-offs set balance rejection rates. metrics, including AUROC, ROC, PR curves used assess performance. study concludes by highlighting benefits while acknowledging its limitations. proposed leverages state-of-the-art techniques draws inspiration cutting-edge methodologies.

Language: Английский

Citations

3
Prediction and assessment of credit risk using an adaptive Binarized spiking marine predators’ neural network in financial sector DOI

Vadipina Amarnadh,

Nageswara Rao Moparthi

Multimedia Tools and Applications, Journal Year: 2023, Volume and Issue: 83(16), P. 48761 - 48797

Published: Nov. 3, 2023

Language: Английский

Citations

3
Smart Financial Investor’s Risk Prediction System Using Mobile Edge Computing DOI
Caijun Cheng, Huazhen Huang

Journal of Grid Computing, Journal Year: 2023, Volume and Issue: 21(4)

Published: Dec. 1, 2023

Language: Английский

Citations

3