Lahore Garrison University Research Journal of Computer Science and Information Technology
Not a member yet
227 research outputs found
Sort by
Unmasking the Deception: A Focused Survey of Machine Learning Techniques for Fake Review Detection
No full textThe expansion of false reviews on e-commerce sites turns out to be a significant issue for the integrity of user-generated content. False reviews may influence the customer’s choice and trustworthiness of sellers. Such reviews are hard to detect because of the absence of labeled data and evolving strategies of spammers. Subsequently, unsupervised learning has become a scalable solution for researchers to detect fake reviews. The paper constitutes a dedicated literature review of machine learning methods, including clustering, anomaly detection autoencoders and hybrid architectures. Supervised learning, Transformer based and deep learning architectures have also been discussed in the review. Based on this taxonomy and comparative analysis, this paper sheds light on the frequently used datasets, evaluation metrics and key trends. Finally, the review concludes with the discussion of the principal limitations and future directions, wherein the attention is paid to semi-supervised learning, multimodal data integration and more flexible and transparent models. The intent of the review is to act as a source of knowledge for the researchers who intend to contribute to the body of knowledge in the area of fake review identification by utilization of unsupervised learning and hybrid learning schemes
An Explainable Deep Learning Framework for Cardiovascular Risk Prediction Using ECG and Imaging Data
No full textCardiovascular disease (CVD) is now the number one cause of death globally, and for its timely and reliable diagnosis, it is required for the prevention and treatment of CVD. In this work, we propose an explainable deep learning model for CVD risk prediction with 12-lead electrocardiogram (ECG) recordings in the PTB-XL dataset. A 1D-CNN is used to extract features from ECG signals automatically, and this process is further enhanced by XAI methods, such as SHAP and Grad-CAM, to increase interpretability and clinical transparency. Noise filtering, signal segmentation, normalizing and data augmentation enable input pre-processing to be clean and consistent for robust learning. The effectiveness of the proposed model has been validated through holdout and cross-validation procedures, yielding good classification performance. The model achieves high accuracy, precision, recall, and F1-score in distinguishing CVD cases from normal ones. It has provided a reliable and scalable system for real-time cardiac risk estimation and serious decision-making by unifying diagnostic precision and interpretability
Developing MLP based prediction system for anticancer drug response using hybrid features of genomics and cheminformatics
Full text linkTraditional cancer treatment methods have become less effective due to the increasing diversity of cancer types. To address this, precision medicine has gained support within the medical community. This approach tailors treatment to individual patients based on their specific disease characteristics. However, a major challenge lies in accurately predicting how a patient will respond to a specialized drug. Numerous machine learning-based predictive systems have been developed to address this challenge. These systems utilize genomic signatures and the chemical structure of drugs to predict drug activity. In this paper, we introduce a Multi-Layer Perceptron (MLP) based system for predicting the response of anticancer drugs. Our system utilizes hybrid features derived from both genetic expression and the chemical structure of drugs. It is developed using the well-known GDSC dataset (Genomics of Drug Sensitivity in Cancer). Our system achieved a lower Root Mean Square Error (RMSE) value of 0.889, in contrast to the RMSE value of 0.983 obtained by the current state-of-the-art (SOTA) system, SwNet. This indicates superior predictive accuracy. The findings suggest that our proposed research holds promise for the development of targeted drugs for anticancer treatments
MULTI-OBJECTIVE OPTIMIZATION BASED DISTRIBUTED TASK OFFLOADING IN FOG COMPUTING
Full text linkDue to an increasing number of IoT devices, a massive amount of data is generating daily. Initially, Cloud-centric Internet of Things (CIoT) based architecture is used for data processing, storage, and analysis. But it is difficult for the CIoT to handle the huge amount of data produced by these devices. Furthermore, cloud data centers are far away from IoT devices so transmitting data to the cloud will result in more bandwidth consumption, cost, and latency issues that’s why cloud computing is not suitable for real-time applications. To overcome these problems the concept of fog computing is introduced that extends cloud computing by moving the facilities to the edge of the network. Fog computing provides low latency, real-time processing near the edge of the network. Fog computing contains heterogeneous and intrinsically resource-constrained devices. Due to resource-constrained devices at the edge of the network resource management is a necessary issue to efficiently distribute resources and move some tasks to the other entity for execution to balance the load it’s called offloading. Different techniques of task offloading are used in fog computing that’s main purpose is to maximize and effective resource utilization, minimizing latency, cost, and energy consumption. This research is concerned with the design and implementation of a multi-objective optimization-based distributed task offloading algorithm in which we will optimize more than two objectives simultaneously like optimization of performance and resource utilization metrics
Formal Modeling of Evolving Systems using Petri nets Based Vocabulary
No full textIn this research we proposed a Petri nets based vocabulary for the development of self adaptive systems. Self adaptive systems (SAS) are systems that undergo dynamicity by reconfiguring their behaviors in accordance with the system needs for execution of programs. Self adaptation has become building block for developing computer systems as it make systems work with higher precision, validity reliability with minimal human effort. Idea behind self adaptation is to enable systems to overcome runtime issues with no or minimal human intervention. SAS consists of two subsystems, a managed system and a self-adaptive system that works as feedback loop. In our approach we used MAPE-K feedback loop that comprises on components monitor, analyze, plan, execute and knowledge to perform self-adaptation. Formal verification of self adaptive systems can be done by specifying actions to be performed by evolving systems. A model is proposed to formally express self adaptive systems. For Formal specification of Evolving systems we have used Linear Temporal Logic (LTL) for the formalism as LTL provides high level of understandability with less ambiguous nature. Verification of system properties have been done using SPIN model checker environment
Integrating Chatbots In Educational Administration For Improved Language Learning Outcomes
Full text linkThis study investigates using chatbots in educational management to improve language learning outcomes. Due to their ability to offer individualized and practical help to learners, chatbots have been introduced as educational aids. The effect of chatbot integration on language learning results is examined in this study, along with its theoretical and practical ramifications. The study uses a mixed-methods approach, integrating qualitative student experiences and quantitative data on language competence levels. The results show a link between using chatbots and successful language learning. While practical implications emphasize the importance of seamless integration and user-centered design, theoretical implications underline the relevance of technology-mediated interactions in education. In its final section, the paper discusses its limits and makes suggestions for educational institutions looking to use chatbots to improve language learning
Formal modelling and verification of autonomous reasoning based flight simulation system
Full text linkThe emerging trends towards intelligent computing have drastically changed the human lifestyle in every facet of their lives. Due to the swift escalation of smart systems and emerging technologies, human life has become much more dependent and even more addicted to fulfilling their desire using smart and tiny resource-bounded gadgets. This revolution has evolved towards autonomous decision support systems. Autonomous decision support systems have the ability to acquire information autonomously, reason the information, and adapt behavior accordingly. As these systems are deployed in a highly decentralized environment and exhibit complex adaptive behavior, however, the inconsistent nature of information may raise different challenging issues. This paper presents a multi-agent environmentally-aware framework for modeling and reasoning flight management systems. This system has a sound reasoning mechanism to execute and monitor flight control activities while considering liveness and safety-critical constraints. We use the UPPAAL model checker to formally analyze the system’s behavior and verify correctness properties
User-Generated Content Analysis: Classification of Factors Affecting Customer Needs for E-commerceRecommender System
No full textOnline selling and buying have significantly increased since COVID-19. As consumer and potential customer demand for online services increases significantly, more and more businesses are turning to electronic commerce to acquire a competitive edge. Companies must acknowledge and give priority to client needs if they want to compete in the market. While interviews, questionnaires, and observation are no longer effective methods for determining customer needs, as theavailability of user-generated online content (UGC) is increasing due to social networking sites (SNSs), it can be used to determine client needs to develop products or services that meet those needs. UGC has accumulated a wealth of information on people’s beliefs, routines, and experiences. Research on the utilization of UGC for e-commerce business applications involves various challenges and approaches, and few studies have summarized the research work performed till now to get a clear picture. First, the study derives a general framework for summarizing the state-of-the-art research. Second, we categorize research based on the manner UGC can be classified, filtered, and understood. Furthermore, we discuss models and techniques used to identify factors affecting customer needs from this content. Lastly, we identify the challenges and limitations faced in the utilization of the user-generatedcontent. To determine client wants, this study examines and categorizes user-generated content (UGC) from social networking sites and e-commerce websites. The classification of user-generated content based on context-dependent behavior, biased reviews, and minority groups has a significant impact on identifying customer demands. The limitations and difficulties of the current methods, models, and data sources are underlined. From this novel review, the researchers may get a sense of the state of the literature today and can observe the difficulties associated with categorizing reviews to identify needs. No such study has categorized the literature on UGC analysis in this manner
Brain Tumor Classification
No full textThe increasing prevalence of brain tumors, which are abnormal growths that occur inside the brain, is a significant problem in medicine. Brain tumor classification is a difficult task in the field of medical image processing. This is because manual categorization can also lead to incorrect diagnoses and forecasts. There is also the possibility that manually analyzing large volumes of data might be challenging. The foundations of effective treatment are a precise diagnosis and quick action, both of which are essential. This research aims to delve further into the topic of brain tumor classification using several techniques such as Random Forest (RF), Support Vector Machine (SVM), K-Nearest Neighbors (KNN), Logistic Regression (LG), and Convolutional Neural Network (CNN). Principally, this investigation aims to develop a dependable and precise system that can autonomously detect and classify many forms of brain tumors, such as pituitary tumors, gliomas, and meningioma, among others. The investigation utilized a diverse range of brain-derived magnetic resonance imaging datasets. This research evaluates the effectiveness of each algorithm by including performance parameters such as accuracy, precision, recall, and F1 score. This study concludes that CNN exhibits the highest accuracy, scoring 99.8 percent, after comparing the results of each discussed algorithm
HYBRID DEEP LEARNING APPROACH TO IDENTIFY INTRUSION DETECTION WITH IMBALANCE DATASETS
Full text linkThe intrusion detection system is a computer-based system that constantly identifies all types of malicious activities by monitoring the network traffic. These intrusions and doubtful activities disturb all business activities performed over the public network, such as the Internet and all connected networks. It is an essential system to provide consistent and reliable transfer of information to complete e-commerce and e-business transactions and private communication using social sites. Various deep learning techniques are used to identify security attacks by observing the typical system usage profile and to restrict all of the network traffic if it is outside the scope of the standard profile. Our proposed system is used to combine various deep learning techniques to develop a hybrid deep learning model to identify any security attack in the network. The proposed hybrid deep learning model is trained using an integrated and balanced dataset by merging already available imbalanced benchmark datasets such as NSL-KDD, ISCX, CICIDS2017, and UNSWNB15. Our proposed system is limited to identifying security attacks in benchmark datasets and restricted to available deep-learning techniques and algorithms