UMT Artificial Intelligence Review

Roman Urdu to Urdu Machine Transliteration by Using T5 Transformer

2025-12-23T05:41:52+00:00

Transliteration is the process of simply analyzing the words in the resource language to the words in the goal language, without any change in meaning. This method transforms the syntax of a text in resource speech into characters of the target language, known as machine transliteration. Recent studies indicate that no dedicated transliteration machine currently exists that covers the issue of RU-U Machine Translation. Previous researchers have attempted to solve this problem using the deep learning techniques, particularly RNN model. Recurrent Neural Network (RNN) transformers are built to manage sequence input information, like natural language, for tasks like translation and text summarization. This model works better on short sentences than long sentences. In the proposed methodology, T5 transformers are encoder-decoder models that translate NLP issues into text-text format. T5 is a transfer learning and the transformers used in this paper are trained on 101 languages including resources language and after training on our parallel data set which consists of 1,107,156 sentences, the study achieved a remarkable result of 91.56 Blue Score.

An Intelligent Method for Improving Credit Card Fraud Detection Using a Hybrid LSTM and Deep Neural Network Framework

2025-12-29T04:06:22+00:00

E-commerce has caused a great transformation in the chain of operations through which companies all over the world transact their businesses. However, with the rapid increase in online shopping, the prevalence of online fraud, particularly credit card fraud has emerged as one of the major security threats connected with e-commerce. The classical models of fraud detection easily address the problems of imbalanced data, pattern of the poorly-sequentially recorded data, and the need to detect the fraud instantly. To address the challenges mentioned in this study, a hybrid architecture which is a fusion of Long Short-Term Memory (LSTM) unit and Deep Neural Network (DNN) modules is propsoed. The DNN component is meant to discover complex interrelatedness of diverse features. Whereas, the LSTM layer establishes a temporal connection which exists in a series of dealings. The preprocessing stage applies the method of the Synthetic Minority Over-Sampling Technique (SMOTE) to solve the issue of unrepresentative classes. The model is tested on the publicly available credit card frauds dataset. It is observed that the proposed model shows a better performance with 99.6% accuracy, 94.5% precision, recall of 91.2% and ROC-AUC of 97.3%, respectively. The comparative study reveals that the hybrid model is superior to the traditional algorithms, including logistic regression, decision trees, LightGBM, and single-created LSTM models, with regard to prediction performance. The presentation of the confusion matrices, the precision-recall curves, and the learning curves is also used to justify the measures of the soundness of the model and its generalizability, without showing the training and validation loss. To conclude, all of these visual tests confirm the reliability of the system under various conditions of the working environment. On the whole, the study adds significantly to the development of a more efficient and scalable fraud detection system, the overall purpose of which is to enhance the level of safety of virtual transaction setups and employ it to other industrial domains, such as energy.

Impact of Artificial Intelligence on the Efficiency of Human Resource Management Practices: A Human-centered Approach

2026-01-02T06:02:17+00:00

The current study aimed to improve Human Resource Management (HRM) through the introduction of novel technologies. Furthermore, this study primarily focused to scrutinize the scientific studies carried out in recent years regarding the application of Artificial Intelligence (AI) in HRM. In order to enhance theoretical as well as empirical knowledge, the study also examined the evolution of conceptual, social, and intellectual structures within the field of HRM. To achieve the intended objectives, a bibliometric analysis was conducted with the help of VOS viewer and Excel for those articles and reviews indexed in the Scopus database. The study emphasized the analysis of research work carried out during the time period (1999-2024) that investigated AI utilization for HRM. The initial investigation found 761 such documents among which 141 were shortlisted using the PRISMA methodology. It was determined that during the time period (2017-2023), there has been a consistent rise in annual publications with the most notable surge (110.7%) in 2022. The most dominant themes uncovered in the HRM using keyword analysis were information management, resource allocation, and resource management. The main application of AI in HRM revolves around the themes of recruitment, talent management and ethical considerations. With a growing interest in consolidating the existing knowledge, review articles attracted significant attention. Three were the notable contributors with the latter being cited the highest despite authoring fewer publications. The surge in publications in the year 2022 shows that the aftermath of COVID-19 pandemic resulted in increased adoption of AI for efficient HR management. “Human Resource Management Review” was the top journal in this theme, whereas the journal titled “International Journal of Human Resource Management” had the maximum citations. The study had certain limitations caused by potential biases resulting from selection criteria, methodological choices, and dependence on a single Scopus database with possible inaccuracies. To address these issues, future studies should investigate publication frequencies in electronic HRM to understand observed differences and the influence of AI in HRM practices and results. It is pertinent to mention that the Scopus data sheet had various inaccuracies, such as misclassifying review articles as regular articles, labelling regular articles as reviews, and errors in publication years. Researchers are advised to verify datasets with original documents and encourage Scopus to enhance data accuracy in order to uphold the credibility of research findings.

Evaluation Evaluation of Search Engines Performance on Standardize Queries

2025-12-23T06:13:03+00:00

The rapid expansion of web content has intensified the need for efficient and accurate search engines capable of responding effectively to diverse user queries. The study aims to compare and evaluate the performance of five major search engines (Google, Bing, Yahoo, Baidu, and DuckDuckGo) in terms of their retrieval speed and relevance across different query types: navigational, informational, and transactional. A set of standardized queries was designed, categorized, and submitted to each search engine under controlled experimental conditions. The number of relevant documents retrieved and the retrieval speed were recorded and analyzed. The results showed that Bing outperformed others in navigational queries, Google dominated informational queries, and Yahoo surprisingly excelled in transactional searches. DuckDuckGo demonstrated competitive performance, especially for navigational tasks. While, Baidu showed strength mainly in informational queries. The findings confirmed that no single search engine is superior across all query types and that search engines optimize differently depending on user intent. This research highlights the importance of query-type-specific evaluation for a better understanding of search engine performance and user satisfaction.

Mitigating Mitigating Risks in Cloud Accounting: Strategies and Implications for Financial Institutions

2025-12-23T06:19:28+00:00

This study investigates the impact of cloud accounting technology on businesses in Pakistan with a focus on the banking sector. As technological advancement accelerates, cloud accounting has become essential to maintain competitiveness. The current research aims to explore the adoption, benefits, challenges, and risks associated with cloud-based accounting systems. A deductive methodology was used. Data was collected from 200 experienced bank employees through questionnaire. A cross-sectional analysis was conducted to examine key variables, such as relative advantage, complexity, and compatibility. Quantitative analysis techniques were applied to identify the patterns and relationships. The findings revealed that while cloud accounting offers significant benefits—such as cost efficiency, scalability, and operational flexibility—it also introduces new risks, including data security breaches and compliance concerns. The study proposes effective risk mitigation strategies, including strong encryption protocols, continuous system monitoring, and rigorous vendor vetting. These insights underscore the importance of proactive risk management and collaboration among stakeholders to ensure data integrity and regulatory compliance in cloud environment.