Current Issue

All articles published in this issue have undergone a thorough peer review process, and stringent checks for repetition rates have been implemented to ensure the integrity of the content.

Total number of articles in this issue: 8
Total number of pages in this issue: 0

For inquiries regarding the content of specific articles, please feel free to contact the respective authors via their provided email addresses. For questions related to the journal itself, please reach out directly to SUAS Press.

Year

2024

Volume

1

Number

1

Status

Archived

Published

2024 October 10

Articles

Open Access Research ArticlePublished 10 October 2024

Gradient Descent Algorithm Optimization and its Application in Linear Regression Model

This paper systematically analyzes the application of gradient descent algorithm in linear regression model and proposes a variety of optimization methods. The basic concepts and mathematical expressions of linear regression model are introduced, and the basic principles and mathematical derivation of gradient descent algorithm are explained. The specific application of gradient descent algorithm in parameter estimation and model optimization is discussed, especially in big data environment. Several optimization methods of gradient descent algorithm are proposed, including learning rate adjustment, momentum method, RMSProp and Adam optimization algorithm. This paper discusses the advantages and disadvantages of gradient descent algorithm and its challenges in practical application, and proposes future research directions. The research results show that the improved gradient descent algorithm has higher computational efficiency and better convergence when processing large-scale data and complex models.

DOI10.5281/zenodo.13753916 ARKark:/40704/AJNS.v1n1a01
View ArticleScholar

Open Access Research ArticlePublished 10 October 2024

Weight Initialization Methods in Convolutional Neural Networks and Their Impact on Training Efficiency

This study analyzes the application effects of different weight initialization methods in convolutional neural networks, focusing on the impact of these methods on training efficiency. Through analysis and derivation from existing studies, we expect significant differences in training speed, stability, and final performance among different initialization methods. Specifically, this study compared random initialization, Xavier initialization, He initialization, LeCun initialization, and Orthogonal initialization. The results show that choosing an appropriate weight initialization method can significantly improve the training efficiency and performance of the model. The main contribution of this article is to systematically compare several common weight initialization methods and propose performance expectations based on existing research to provide a reference for subsequent research. Future research can verify these expected results through actual experiments and explore novel initialization strategies to further improve the training efficiency and performance of the model.

DOI10.5281/zenodo.13755632 ARKark:/40704/AJNS.v1n1a02
View ArticleScholar

Open Access Research ArticlePublished 10 October 2024

Research on the Application of Deep Learning-Based Text Classification Algorithms in Data Mining

This article explores the application of text classification algorithms based on deep learning in data mining. By reviewing the existing literature, the application of convolutional neural network (CNN), recurrent neural network (RNN) and transformer (Transformer) in text classification is introduced, and related research on data mining and analysis is analyzed. The experiment used the IMDB movie review data set to verify the effectiveness of the deep learning model through data preprocessing, feature extraction and model training. The experimental results show that the model's accuracy on the test set is 86%, the precision rate is 85%, the recall rate is 88%, and the F1 value is 86%. Research shows that deep learning models can significantly improve text classification performance. This article also discusses the significance, limitations and future research directions of the research findings, including improvements in data acquisition and annotation, computing resource optimization, and model interpretability.

DOI10.5281/zenodo.13762126 ARKark:/40704/AJNS.v1n1a03
View ArticleScholar

Open Access Research ArticlePublished 10 October 2024

Using Data Visualization Technology to Improve the Effectiveness of Township Environmental Management

This study discusses the application of data visualization technology in improving the effectiveness of township environmental management. Through literature review, actual case analysis and experimental verification, the important role of data visualization technology in environmental monitoring, data analysis and decision support is revealed. The results show that data visualization technology can not only improve the accuracy and comprehensibility of environmental data, but also enhance the scientificity and transparency of environmental management decisions and improve emergency response capabilities. This paper first introduces the research background and significance, and then reviews the development history of data visualization technology and its current application in environmental management. Then, the specific application of data visualization technology in township environmental management is analyzed in detail, including data collection and processing, the selection of visualization tools and technologies, and actual application cases. Finally, through experiments and result analysis, the application effect of data visualization technology is evaluated, and future research directions are proposed.

DOI10.5281/zenodo.13884870 ARKark:/40704/AJNS.v1n1a04
View ArticleScholar

Open Access Research ArticlePublished 10 October 2024

A Dynamic Credit Risk Assessment Model Based on Deep Reinforcement Learning

This study presents a new credit risk assessment model based on deep learning (DRL), addressing the limitations of static models in adapting to financial changes. The model leverages the Deep Q-Network architecture developed with the importance of replication, DQN distribution, and ring noise to capture the correlations and non-linearities in the credit data and transition to financial reform. An analysis of the extensive data of 1,250,000 credit applications shows the best performance model, achieving an AUC-ROC of 0.92 and a 15% improvement in the ratio for state-of-the-art machine learning methods. The DRL model exhibits remarkable adaptability in simulated dynamic scenarios, maintaining an average AUC-ROC of 0.89 across various economic shocks and market shifts. An ablation study reveals the significant contributions of each model component, with prioritized experience replay emerging as the most impactful for adaptability. The proposed approach offers promising implications for enhancing credit risk management practices, potentially enabling more inclusive lending while mitigating default risks. However, challenges remain in model interpretability and potential bias perpetuation, necessitating further research into transparent and fair AI-driven credit assessment systems. This study contributes to the growing body of research applying advanced machine-learning techniques to critical financial decision-making processes.

DOI10.5281/zenodo.13905241ARKark:/40704/AJNS.v1n1a05
View ArticleScholar

Open Access Research ArticlePublished 10 October 2024

A Review of Feature Engineering Methods in Regression Problems

This paper comprehensively reviews the importance of feature engineering in regression problems and its existing methods. Different feature processing techniques, including data transformation, interactive feature generation, dimensionality reduction, and technical indicators, are analyzed, and some existing research results are combined to discuss how these techniques affect the predictive ability of the model. By systematically summarizing the advantages and disadvantages of various methods, this review provides researchers and practitioners with a comprehensive understanding of existing feature engineering techniques. In addition, this paper also explores the existing research gaps and possible future development directions, and proposes more application scenarios and research topics for related technologies.

DOI10.5281/zenodo.13905622ARKark:/40704/AJNS.v1n1a06
View ArticleScholar

Open Access Research ArticlePublished 10 October 2024

Hybrid Mixed Integer Programming with Machine Learning for Digital Asset Management

Solving Mixed Integer Programming (MIP) problems is critical for numerous applications in digital asset management. Traditional solvers, which rely on techniques like branch-and-bound and branch-and-cut, face computational challenges when dealing with large-scale and complex problem instances. Recent advances in machine learning (ML) have introduced new approaches to enhance these traditional solvers by optimizing key decision-making processes such as branching, cut generation, and heuristic selection. In this paper, we examine hybrid methods that combine ML with classical MIP methodologies, leveraging supervised learning, reinforcement learning, and graph neural networks to improve solver efficiency and scalability. These hybrid methods significantly reduce computational overhead by making smarter decisions during the optimization process, resulting in faster convergence and higher-quality solutions. We also discuss the challenges of generalization and integration of ML models with existing solvers and propose future research directions to further advance this field.

DOI10.5281/zenodo.13909877ARKark:/40704/AJNS.v1n1a07
View ArticleScholar

Open Access Research ArticlePublished 10 October 2024

Multimodal Sentiment Analysis: A Study on Emotion Understanding and Classification by Integrating Text and Images

The advent of social media and the proliferation of multimodal content have led to the growing importance of understanding sentiment in both text and images. Traditional sentiment analysis relies heavily on textual data, but recent trends indicate that integrating visual information can significantly improve sentiment prediction accuracy. This paper explores multimodal sentiment analysis, specifically focusing on emotion understanding and classification by integrating textual and image-based features. We review existing approaches, develop a hybrid deep learning model utilizing attention mechanisms and transformer architectures for multimodal sentiment classification, and evaluate its performance on benchmark datasets, including Twitter and Instagram data. Our findings suggest that multimodal approaches outperform text-only models, especially in more nuanced sentiment cases such as sarcasm, irony, or mixed emotions. Moreover, we address key challenges like feature fusion, domain adaptation, and the contextual alignment of visual and textual information. The results provide insights into optimizing multimodal fusion techniques to enhance real-world application performance.

DOI10.5281/zenodo.13909963 ARKark:/40704/AJNS.v1n1a08
View ArticleScholar

Copyright © 2026 The Author(s). Published by Southern United Academy of Sciences.
This work is licensed under a Creative Commons Attribution 4.0 International License , which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Academic Journal of Natural Science (AJNS)
Published by Southern United Academy of Sciences Limited ISNI: 0000000512776460, operated by the Publications Division.