Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Lubna Abdulaziz Alharbi has a total of 39 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    DeepRawNet: empowering deepfake audio detection through dynamic enhancements

    by Lubna A. Alharbi, Jasim Alnahas, Ali Javed, A’AEshah Alhakamy, Marriam Nawaz, Hafiz Malik, Hafsa Ilyas, Faris Alfaifi

    Background: The generation of deepfake audio poses significant challenges to the reliability and security of automatic speaker verification (ASV)-based systems. ASV systems having applications in fintech, surveillance, home automation, security, etc ., are susceptible to a variety of deepfake/voice cloning attacks, including speech synthesis and voice conversion (VC). Impostors launch audio deepfake attacks on ASV systems to compromise their security and cause financial losses, data breaches, etc . Methods: To combat such threats, we propose a robust and generalized audio deepfakes detection framework, DeepRawNet, by processing raw audio waveforms. Specifically, DeepRawNet is the enhanced version of RawNet2, introduces three key innovations: (1) we employ the Parametric Rectified Linear Unit (PReLU) activation in the residual blocks over Leaky ReLU in the RawNet2, introducing a learnable negative slope to enhance adaptive feature extraction, (2) substituting the simple convolution layer with a transpose convolution layer in the residual block addresses downsampling issues while preserving fine-grained temporal information crucial for capturing complex patterns in raw audio, (3) we incorporate the LogSoftmax activation function to stabilize and optimize learning during training and inference. These architectural refinements empower our DeepRawNet model with improved adaptability, robust learning capabilities, and enhanced capacity to capture complex temporal dependencies and discriminative patterns in the audio, making it a more effective solution for audio deepfake detection. Results: We performed a rigorous evaluation of our proposed method on ASVspoof2019-LA and ASVspoof2021-LA/DF datasets, including algorithm-wise and cross-corpora evaluation, an ablation study with different model configurations, and comparison against baseline models and existing approaches. Experimental results highlight the improved performance of DeepRawNet against the ASVspoof baselines, improved generalization across diverse spoofing attacks, particularly for the most challenging VC attacks, and effectiveness in combating deepfake audio threats.


  • Smart agriculture: IoT-Based smart irrigation with advanced fuzzy logic control

    by Anas Bushnag, Slim Ben Chaabane, Rafika Harrabi, Lubna A. Alharbi, Maha Alshmrani, Shakour Abuzneid


  • Open Access

    Heart Disease Prediction of Cleveland Clinic Patients using Advanced Machine Learning Algorithms

    by Lubna A. Alharbi

    Globally, cardiovascular diseases (CVDs) constitute the primary cause of morbidity and mortality worldwide. Early diagnosis of those at risk of CVDs may lower the number of avoidable fatalities. It has been shown that machine learning (ML) is helpful in anticipating cardiac issues. Adoption of a prediction system that can detect cardiac diseases before they deteriorate would offer people worldwide enormous hope and help in decision-making. ML has become a popular technique for generating predictions from enormous real-world datasets. It has also been discovered that many ML classifiers contain issues and flaws. However, the latest ML algorithm from the boosting family, i.e., XGBoost, may enhance performance and assist in exact prediction. As a result, this study will compare XGBoost to other prominent classifiers in terms of their capacity to anticipate and improve performance. ML classifiers, such as Multilayer Perception, K-nearest neighbours (K-NN), Support Vector Classifier, CART and XGBoost algorithms, are used to differentiate between healthy and CVD patients. When compared to competing classifiers, the XGB classifier achieves 89% accuracy, 87% precision, 94% sensitivity, 94% specificity, 90.2% F1 score, 81.2 % ROC, 78.90% Mathew coefficient and 3.87% log loss. In the future other ML classifiers such as Random Forest, Multilayer Perceptron, K-nearest Neighbor, Extra Tree Classifier, Extreme Gradient Boosting, Support Vector classifier, Stochastic Gradient Descent, AdaBoost, Classification and Regression Tree and Gradient Boosting along with these algorithms could be applied to compare the MLA efficiency.


  • Open Access

    Artificial Rabbits Optimizer With Machine Learning Based Emergency Department Monitoring and Medical Data Classification at KSA Hospitals

    by Lubna A. Alharbi

    The Emergency Departments (EDs) in health centres located in the main areas of Saudi Arabia have heavy patient inflow because of the pandemic, viral infections, and even on some special occasions like Umrah or Hajj, where pilgrims who travel from one place to another with serious disorders. Other than the EDs, it was important to observe the patient’s activities from ED to other wards in the region or the hospital to track the spread of viral diseases. In this case, deep learning (DL) and machine learning (ML) methods have been used to track the target audience and classify the data into many classes. With this motivation, this study develops an artificial rabbit optimization with a machine learning-based healthcare data classification (AROML-HDC) technique for EDs. The AROML-HDC technique monitors and tracks the patient visit data, treatment given, and length of stay (LOS). In addition, the AROML-HDC technique designs an effective ARO algorithm for the optimal selection of feature subsets. Next, the class-specific cost regulation extreme learning machine (CSCR-ELM) classifier is applied for effective medical data classification. Finally, the grasshopper optimization algorithm (GOA) was used to adjust the parameters related to the CSCR-ELM classifier. The experimental outcome of the AROML-HDC approach is tested on the benchmark Cleveland dataset and the Statlog dataset comprising 297 and 270 samples, respectively. The simulation results signify the improved performance of the AROML-HDC technique over other recent methods with maximum accuracy of 93.22% and 94.05% Cleveland dataset and the Statlog dataset, respectively.


  • Optimal trajectory UAV path design based on bezier curves with multi-hop cluster selection in wireless networks

    by Likun Li, Yinsheng Fu, Kun Yu, Ahmed M. Alwakeel, Lubna A. Alharbi

    In the past decade, multi-UAV systems have gained significant attention and companies from in various fields. The main aims of growing such systems are able to operate in a coordinated way in complicated situations. To encompass and increase the speed of the applications, this research proposed optimization with intelligence. In the existing work, it does not detect the collision effectively. In this work, the Particle Swarm Optimization algorithm based on fast matching square is proposed for flightpath planning in Multi-UAVs is used. A distributed full coverage optimal route design is derived using the particle swarm optimization (PSO) technique, and a trajectory planner is built utilizing a dynamic fitness function. We applied the PSO technique for every UAV individually to achieve active fitness, maximizing the fitness function while reducing the cost function. The novelty of this work is using the dynamic fitness function for PSO while selecting optimal data. Fast matching square FM2 helps to detect the collision and avoid collisions between the paths. The suggested method generates an optimal path for multi-UAVs and updates the maps for swarms of UAVs. According to experimental findings, the suggested PSO-FM2 outperforms the previous FOA in aspects of convergence and accuracy. Finally, the proposed method is compared with existing algorithms such as MSFOA and PSO.


  • Open Access

    Cloud Servers: Resource Optimization Using Different Energy Saving Techniques

    by Mohammad Hijji, Bilal Ahmad, Gulzar Alam, Ahmed Alwakeel, Mohammed Alwakeel, Lubna Abdulaziz Alharbi, Ahd Aljarf, Muhammad Umair Khan

    Currently, researchers are working to contribute to the emerging fields of cloud computing, edge computing, and distributed systems. The major area of interest is to examine and understand their performance. The major globally leading companies, such as Google, Amazon, ONLIVE, Giaki, and eBay, are truly concerned about the impact of energy consumption. These cloud computing companies use huge data centers, consisting of virtual computers that are positioned worldwide and necessitate exceptionally high-power costs to preserve. The increased requirement for energy consumption in IT firms has posed many challenges for cloud computing companies pertinent to power expenses. Energy utilization is reliant upon numerous aspects, for example, the service level agreement, techniques for choosing the virtual machine, the applied optimization strategies and policies, and kinds of workload. The present paper tries to provide an answer to challenges related to energy-saving through the assistance of both dynamic voltage and frequency scaling techniques for gaming data centers. Also, to evaluate both the dynamic voltage and frequency scaling techniques compared to non-power-aware and static threshold detection techniques. The findings will facilitate service suppliers in how to encounter the quality of service and experience limitations by fulfilling the service level agreements. For this purpose, the CloudSim platform is applied for the application of a situation in which game traces are employed as a workload for analyzing the procedure. The findings evidenced that an assortment of good quality techniques can benefit gaming servers to conserve energy expenditures and sustain the best quality of service for consumers located universally. The originality of this research presents a prospect to examine which procedure performs good (for example, dynamic, static, or non-power aware). The findings validate that less energy is utilized by applying a dynamic voltage and frequency method along with fewer service level agreement violations, and better quality of service and experience, in contrast with static threshold consolidation or non-power aware technique.


  • Open Access

    Artificial Jellyfish Optimization with Deep-Learning-Driven Decision Support System for Energy Management in Smart Cities

    by A. Al-Qarafi, Hadeel Alsolai, Jaber S. Alzahrani, Noha Negm, Lubna A. Alharbi, Mesfer Al Duhayyim, Heba Mohsen, M. Al-Shabi, Fahd N. Al-Wesabi

    A smart city is a sustainable and effectual urban center which offers a maximal quality of life to its inhabitants with the optimal management of their resources. Energy management is the most difficult problem in such urban centers because of the difficulty of energy models and their important role. The recent developments of machine learning (ML) and deep learning (DL) models pave the way to design effective energy management schemes. In this respect, this study introduces an artificial jellyfish optimization with deep learning-driven decision support system (AJODL-DSSEM) model for energy management in smart cities. The proposed AJODL-DSSEM model predicts the energy in the smart city environment. To do so, the proposed AJODL-DSSEM model primarily performs data preprocessing at the initial stage to normalize the data. Besides, the AJODL-DSSEM model involves the attention-based convolutional neural network-bidirectional long short-term memory (CNN-ABLSTM) model for the prediction of energy. For the hyperparameter tuning of the CNN-ABLSTM model, the AJO algorithm was applied. The experimental validation of the proposed AJODL-DSSEM model was tested using two open-access datasets, namely the IHEPC and ISO-NE datasets. The comparative study reported the improved outcomes of the AJODL-DSSEM model over recent approaches.


  • Top Cited Papers
    Open Access

    An Efficient Approach for Crops Pests Recognition and Classification Based on Novel DeepPestNet Deep Learning Model

    by Naeem Ullah, Javed Ali Khan, Lubna Abdulaziz Alharbi, Asaf Raza, Wahab Khan, Ijaz Ahmad

    Crop pests are to blame for significant economic, social, and environmental losses worldwide. Various pests have different control strategies, and precisely identifying pests has become crucial to pest control and is a significant difficulty in agriculture. Many agricultural professionals are interested in deep learning (DL) models since they have shown significant promise in image recognition. Pest identification approaches in literature have relatively low accuracy in pest recognition and classification due to the complexity of their algorithms and limited data availability. Misclassification of insect pests sometimes leads to using the wrong pesticides, causing harm to agricultural yields and the surrounding environment. It necessitates developing an automated system capable of more accurate pest identification and classification. This paper presents a novel end-to-end DeepPestNet framework for pest recognition and classification. The proposed model has 11 learnable layers, including eight convolutional and three fully connected (FC) layers. We used image rotations techniques to increase the size of the dataset and image augmentations techniques to test the generalizability of the proposed DeepPestNet approach. We used the popular Deng’s crops data set to assess the proposed DeepPestNet framework. We used the proposed method to recognize and classify crop pests into 10-class pests, i.e., Locusta migratoria, Euproctis pseudoconspersa strand, chrysochus Chinensis, empoasca flavescens, Spodoptera exigua, larva of laspeyresia pomonella, parasa lepida, acrida cinerea, larva of S. exigua, and L.pomonella types of insects pests. The proposed method achieved optimal accuracy of 100%. We compared the proposed DeepPestNet approach with traditional pre-trained deep learning (DL) models. To verify the general adaptability of this model, we tested the proposed model on the standard Kaggle dataset “Pest Dataset” to recognize nine types of pests: aphids, armyworm, beetle, bollworm, grasshopper, mites, mosquito, sawfly, and stem borer and achieved an accuracy of 98.92%. The proposed model can provide specialists and farmers with immediate and effective aid in recognizing pests, potentially reducing economic and crop yield losses.