Object detection is a fundamental visual recognition problem in computer vision that has also drawn interest from researchers in recent years. Visual object detection techniques are known to predict labels for object instances and identify the class of a particular object based on its location within the image. It can identify objects in complicated photos and films. Methods based on object detection are necessary for the majority of computer vision-based systems. Ongoing study in the field is accelerating advancement in a number of ways. This paper offers an extensive overview of contemporary object detection methods, classifying them into traditional machine learning, deep learning approaches and recently used transformer based models, while particularly emphasizing the latter’s distinction between two-stage and one-stage detectors. We assess the effectiveness of these techniques by reviewing crucial evaluation measurements, such as precision, recall, and mAP, on standard datasets like COCO and PASCAL VOC. Additionally, we investigate the use of these methods across various fields, emphasizing their effectiveness in tackling practical issues. In conclusion, we recognize and analyze significant obstacles that existing detection models encounter, including the management of small objects and enhancing real-time efficiency, and suggest possible future research pathways intended to progress the cutting-edge in object detection.
In Peer-to-Peer (P2P) networks financial risk modelling is a rapidly growing area in finance. It offers innovative borrowing options however, also presents new challenges. Financial models are essential for assessing borrower loan eligibility. This study integrates network centrality metrics with machine learning techniques to predict the risk of borrower defaults. It uses Mutual Information (MI) scoring to evaluate feature significance and Principal Component Analysis (PCA) for feature extraction. Key predictors include principal balance, expected return, and loss given default. The six Machine Learning (ML) algorithms: Support Vector Machine (SVM), Naïve Bayes (NB), Decision Tree (DT), Logistic Regression (LR), and ensemble methods like Random Forest (RF) and XGBoost are analysed using the Bondora European dataset. The models are evaluated using AUC score, recall, accuracy, precision, and F1 score. The results show that the SVM model provides the best performance, achieving 83
This study investigates market behavior by separately examining investor convergence of six major stock markets from five countries and its relationship with global equity movements using the cross-sectional absolute deviation (CSAD) framework. Traditional asset-pricing theories focus primarily on rational expectations as the foundation of international financial integration; however, they often overlook the influence of irrational expectations and behavioral biases that shape stock market dynamics. To address this gap, the study analyzes herding tendencies of six major stock indices representing five key economies India, Japan, China, Europe, and the United States-using daily index data spanning from April 1, 2015, to March 31, 2024. The methodology places particular emphasis on extreme market conditions and the COVID-19 period to assess how herding behavior may shift during periods of heightened uncertainty and market stress. The findings indicate no evidence of herding during extreme upward or downward market movements, including during the COVID-19 crisis itself. However, when global market returns are considered, herding behavior becomes evident in the US (NASDAQ), Europe, and India, while no such tendencies are observed in the US (NYSE), Japan, or China. Interestingly, during the pandemic period, herding based on global market influences persists only in the European and Indian markets, suggesting heterogeneous behavioral responses in different economies. These insights contribute to asset-pricing literature and offer practical value for investors, traders, analysts, fund managers, and policymakers by enhancing understanding of market efficiency and behavioral risks. The study’s comprehensive dataset, which incorporates six major global markets and covers the COVID-19 crisis period, provides a unique and timely contribution to behavioral finance by offering deeper empirical evidence on herding during a significant global disruption.
Autonomous software testing leverages Artificial Intelligence (AI) and Machine Learning (ML) techniques to create, manage, and execute test cases with minimal human intervention. The objective is to enhance the productivity of testing activities through intelligent algorithms, enabling testers to focus primarily on strategic aspects only. Industries are increasingly making large-scale investments to promote autonomous software testing. Despite its growing significance, the field remains fragmented, with research dispersed across domains such as machine learning, model-based testing, reinforcement learning, and search-based techniques. This fragmentation creates challenges for both practitioners and researchers in developing a cohesive understanding of the state of the art. This systematic literature review consolidates existing approaches to Autonomous Software Testing by identifying, classifying, and synthesizing key contributions across academic research. It further highlights current gaps and challenges, thereby guiding future research directions. This study explores the key phases of software testing where intelligent solutions are most widely applied and analyses their associated trade-offs. It provides a comprehensive overview of AI and machine learning techniques employed in autonomous testing, with findings systematically presented through tables, graphs, and figures to ensure clarity and accessibility. The review is intended to serve as a valuable resource for newcomers to software testing, offering a holistic perspective on the autonomous testing landscape.
In the recent trend of the rise of several software development organizations, there are very few people left who can write fresh code from scratch. If the client does not specify any requirements, the developers will look for reusable code to complete the project. This process of plagiarism is known as code clone. For several reasons, it has become apparent that the amount of code being copied is increasing, and thus, the issues associated with the use of legacy code are increasing. In this paper, we have present an algorithm that takes two programs as input, converts them into a graph representation, and matches subgraphs between them to find code clones. In order to understand the workings of the algorithm, we have taken the help of four program-based case studies and applied our algorithm to find the code clones present in them. Also, a detailed comparison of the developed method with existing techniques.
