SFSaima Farhan

Saima Farhan

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Saima Farhan has a total of 40 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Investigation and Analysis of the Role of Leaf’s Vein Features for Automated Classification of Plants

    by Saima Farhan, Aisha Riaz, Fatima Anjum, Ayesha Iqbal, Faiza Kalsoom

    Several researchers have proposed various methods to classify plants. Veins pattern of leaf contains imperative information for plants identification regardless its complex modality. This research describes a methodology for classifying different types of plants using various features i.e. leaf shape, texture, color and newly proposed vein features. The proposed algorithm comprises of three distinct stages that are preprocessing, feature extraction and classification. The characteristics of plant can be determined from leaf structures. The classification approach used in this research based on plant leaves as leave belongs to different plants has different characteristics. The used dataset for the proposed approach is a flavia database that contains 32 kinds of plant leaves. It is observed from experimental results that the classification accuracy of 97.25% was achieved using a Random Forest classifier when a combination of shape, HSV color moment, texture and venation features are used.

  • Breast Cancer Classification Using Gene Expression Profile Data with a Deep Learning Framework

    by Maria Nazeer, Saima Farhan, Rabia Javed, Ayesha Iqbal, Fatima Anjum

    Breast cancer is a harmful disease that is dangerous to human life. It causes a continuous increase in the overall death rate. It is a challenging task for machine learning models to classify gene expression profile data because of its complex nature. These machine-learning models took a lot of time and consumed larger data during their training, which resulted in inconsistent accuracy. The main purpose of this research is to build a deep neural network model that can accurately classify cancerous and non-cancerous patients by using gene expression profile datasets of breast cancer. This study presents a novel deep learning framework specifically designed for the classification of breast cancer using gene expression profile data. Unlike conventional machine learning approaches that require extensive preprocessing and manual feature extraction, the proposed model automatically learns discriminative features from high-dimensional genomic data. The proposed research implies a highly accurate model having precision (0.91), recall (1.00), and F1 score (0.95) with 95% accuracy. The macro average and weighted average both indicate that the model performs well on average also which is helpful for medical practitioners for early breast cancer prognosis.

  • Open Access

    Enhancing the early detection of Alzheimer’s disease using an integrated CNN-LSTM framework: A robust approach for fMRI-based multi-stage classification

    by Saima Farhan, Yasin Ul Haq, Momina Abdul Khaliq, Seemal Afza, Fahad Ahmad, Tariq Mahmood, Amjad Rehman

    Alzheimer’s Disease poses a significant challenge as a progressive and irreversible neurological condition striking the elderly population. Its incurable nature correlates with a significant rise in death rates. However, early detection can slow its progression and facilitate prompt intervention, thereby mitigating mortality risks. Functional Magnetic Resonance Imaging (fMRI) provides valuable insights into the functional changes within distinct brain regions associated with the disease. The recent research efforts have extracted functional connectivity measures for the classification. These handcrafted functional connectivity features are usually not robust and are computationally intensive. To address the issue, this study introduces an integrated deep-learning framework based on CNN and LSTM networks. This framework autonomously learns both intra-volume and inter-volume features critical for classification tasks. CNNs facilitate feature extraction, while LSTM networks govern the selection of significant features for classification. The key aim of this study is to classify Alzheimer’s disease and its prodromal stage, Mild Cognitive Impairment (MCI). MCI is further categorized as early MCI (EMCI) and late MCI (LMCI). We have evaluated the framework in three dimensions, binary classification, multi-class classification with 3-classes, and multi-class classification with 4-classes. For each dimension, multiple classifications were performed. The results depict the proposed CNN-LSTM framework to attain 99% accuracy and 100% average area under the curve for the majority of the classification.


  • Enhancing network security: an intrusion detection system using residual network-based convolutional neural network

    by Saima Farhan, Jovaria Mubashir, Yasin Ul Haq, Tariq Mahmood, Amjad Rehman


  • Open Access

    Enhancing Voice Spoofing Detection: A Hybrid Approach With VGGish-LSTM Model for Improved Security in Automatic Speaker Verification Systems

    by Komal Shahzad, Saima Farhan, Yasin- Ul- Haq, Rabia Sana, Salman Pathan

    Cybercrimes frequently manifest through the utilization of the internet. Spoofing attacks epitomize an advanced facet of contemporary cybercrime. Considerable research has been undertaken in the domain of these diverse spoofing attack techniques. Recent spoofing attacks predominantly concentrate on voice spoofing, driven by the vulnerabilities of Automatic Speaker Verification systems. Such type of attacks contains significant threats, resulting in fraudulent activities including imitation of someone’s voice for financial scam, bypassing security locks, or getting sensitive information and data. To address these concerns, this research focuses on the detection of voice spoofing attacks by utilizing a hybrid VGGish-LSTM model. While previous transfer learning methodologies have essentially been applied to image classification tasks, our methodology focuses on voice classification. In comparison to VGGish, another transfer learning model i.e. YAMNet is also utilized. These models support the extraction of significant features i.e. embeddings, capable of mitigating bias and facilitating analysis with limited datasets. For classification, LSTM and 1D CNN models are applied on the extracted embeddings. The evaluation uncovers that the hybrid VGGish-LSTM model exhibits better performance than alternative configurations. We achieved a remarkable accuracy of 99.11% with an equal error rate of 1.18% on ASV Spoof 2019 LA dataset. This research helps the general public to recognize spoof attacks, thereby helping in the prevention of frauds by the detection of synthesized speech.


  • Open Access

    Enhancing rice yield prediction: a deep fusion model integrating ResNet50-LSTM with multi source data

    by Aqsa Aslam, Saima Farhan

    Rice production is pivotal for ensuring global food security. In Pakistan, rice is not only the dominant Kharif crop but also a significant export commodity that significantly impacts the state’s economy. However, Pakistan faces challenges such as abrupt climate change and the COVID-19 pandemic, which affect rice production and underscore the need for predictive models for informed decisions aimed at improving productivity and ultimately the state’s economy. This article presents an innovative deep learning-based hybrid predictive model, ResNet50-LSTM, designed to forecast rice yields in the Gujranwala district, Pakistan, utilizing multi-modal data. The model incorporates MODIS satellite imagery capturing EVI, LAI, and FPAR indices along with meteorological and soil data. Google Earth Engine is used for the collection and preprocessing of satellite imagery, where the preprocessing steps involve data filtering, applying region geometry, interpolation, and aggregation. These preprocessing steps were applied manually on meteorological and soil data. Following feature extraction from the imagery data using ResNet50, the three LSTM model configurations are presented with distinct layer architectures. The findings of this study exhibit that the model configuration featuring two LSTM layers with interconnected cells outperforms other proposed configurations in terms of prediction performance. Analysis of various feature combinations reveals that the selected feature set (EVI, FPAR, climate, and soil variables) yields highly accurate results with an R2 = 0.9903, RMSE = 0.1854, MAPE = 0.62%, MAE = 0.1384, MRE = 0.0062, and Willmott’s index of agreement = 0.9536. Moreover, the combination of EVI and FPAR is identified as particularly effective. Our findings revealed the potential of our framework for globally estimating crop yields through the utilization of publicly available multi-source data.


  • Open Access

    Automatic Sarcasm Detection on Cross-Platform Social Media Datasets: A GLoVe and Bi-LSTM Based Approach 

    by Saima Farhan, Rubiya Shoukat, Aqsa Aslam

    Sarcastic remarks on social media platforms have become commonplace, with people expressing their bad feelings in a quite positive manner or in a mocking way. This contradictive nature of sarcasm makes its detection a very challenging task. Many researchers have provided their solutions to perform automatic sarcasm detection from a single domain dataset. Most of them have considered only the content of the text and has ignored the context of the text. Understanding that the context of a text is the most important factor in determining either it is sarcastic or not. This study aims to detect sarcastic remarks from multi-domain dataset by using Bi-LSTM model employed with pre-trained GloVe word embeddings because the GloVe embeddings and Bi-LSTM both are good at capturing contextual information from the provided data. The dataset is generated by the concatenation of three publicly available datasets, the gosh tweets, the news headlines dataset, and the sarcasm corpus v2 dataset. GloVe embeddings will extract contextual and semantic features from the text while the Bi-LSTM model will get trained and tested on those features. The proposed model has achieved 86.35% of accuracy with 88% of Recall, Precision, and F1-score. Different experiments have been done to test the model's reliability. This study's findings indicate that the proposed model yields state-of-the-art or comparable results. The proposed study aids in improving the performance of sentiment analysis. It will also help individuals, and different organizations to identify accurate sentiments of people about individuals or products of an organization.


  • Top Cited Papers
    Open Access

    Intrusion Detection System for Wireless Sensor Networks: A Machine Learning Based Approach

    by Halima Sadia, Saima Farhan, Yasin Ul Haq, Rabia Sana, Tariq Mahmood, Saeed Ali Omer Bahaj, Amjad Rehman Khan

    In this era, there are plenty of wireless devices that are being used with the support of WI-FI (Wireless Fidelity) and they need to be maintained and authorized. Wireless Sensor Networks (WSN), a cornerstone of modern wireless technology, offer cost-efficient solutions for diverse monitoring tasks but are simultaneously exposed to a myriad of security threats, including unauthorized access, attacks, and suspicious activities. These vulnerabilities can significantly degrade the performance and reliability of WSNs, making the early detection and mitigation of such threats imperative. Intrusion Detection Systems (IDS) emerge as crucial tools in safeguarding WSNs against these challenges. Numerous studies focus on enhanced Intrusion Detection model accuracy and decrease in loss with higher Detection Rate and lower False Alarm Rate, because of this, eliminating the repetitive feature of the dataset is exhibited. This study introduces a sophisticated Network Intrusion Detection System (NIDS) designed to safeguard Wi-Fi-based WSNs from prevalent cyber threats, such as impersonation, flooding, and injection attacks. At the heart of our approach is a meticulous feature selection process that enhances the dataset’s utility by eliminating null values, substituting unknown entries with a placeholder (’NONE’), and refining the feature set to include only the most relevant indicators of potential security breaches. Initially, from a pool of 154 features, a subset of 76 is selected, which is further narrowed down to 13 pivotal features, ensuring a focused and efficient analysis. Employing standard scaler function for feature scaling and preprocessing, this research train proposed a Convolutional Neural Network (CNN) based approach aiming for optimal intrusion detection and prevention across multiclass classifications within WSN environments. The study aims to enhance detection accuracy, reduce loss values, and decrease false alarm rates, comparing it to CNN, Deep Neural Network (DNN) (5), DNN (3), and (Long Short-Term Memory) LSTM networks. The model’s performance is evaluated using various metrics, including precision, recall, support, F1 score, and macro-average. The culmination of our research efforts is evidenced by the exceptional performance of the CNN model, achieving an impressive accuracy rate of 97% and a loss metric of 0.14, all while maintaining a minimal False Alarm Rate. This study significantly advances IDS accuracy while simultaneously reducing false alarms, thus fortifying the security posture of WSNs in the face of evolving cyber threats.


  • Open Access

    Convolutional Neural Network-Based Classification of Multiple Retinal Diseases Using Fundus Images

    by Aqsa Aslam, Saima Farhan, Momina Abdul Khaliq, Fatima Anjum, Ayesha Afzaal, Faria Kanwal

    Use of deep learning algorithms for the investigation and analysis of medical images has emerged as a powerful technique. The increase in retinal diseases is alarming as it may lead to permanent blindness if left untreated. Automation of the diagnosis process of retinal diseases not only assists ophthalmologists in correct decision-making but saves time also. Several researchers have worked on automated retinal disease classification but restricted either to hand-crafted feature selection or binary classification. This paper presents a deep learning-based approach for the automated classification of multiple retinal diseases using fundus images. For this research, the data has been collected and combined from three distinct sources. The images are preprocessed for enhancing the details. Six layers of the convolutional neural network (CNN) are used for the automated feature extraction and classification of 20 retinal diseases. It is observed that the results are reliant on the number of classes. For binary classification (healthy vs. unhealthy), up to 100% accuracy has been achieved. When 16 classes are used (treating stages of a disease as a single class), 93.3% accuracy, 92% sensitivity and 93% specificity have been obtained respectively. For 20 classes (treating stages of the disease as separate classes), the accuracy, sensitivity and specificity have dropped to 92.4%, 92% and 92% respectively.


  • Textural and Geometrical Features Based Approach for Identification of Individuals Using Palmprint and Hand Shape Images from Multiple Multimodal Datasets

    by Anum Shaukat, Saima Farhan, Muhammad Abuzar Fahiem, Huma Tauseef, Fahima Tahir, Ghousia Usman

    Identification and classification of biometrics are important research areas in the field of image processing and pattern recognition. Biometrics are the measurement and statistical analysis of physiological and behavioral characteristics of humans. A wide variety of biometric modalities are available, with unimodal biometrics suffering from several factors. The proposed research is novel because it uses a single image of a hand in order to extract a variety of unique characteristics, like hand shape and the palmprint associated with individual hands. Moreover, it obtains higher accuracy with minimum effort. We have chosen the multimodal biometrics, i.e., palmprint and hand shape, from three datasets, i.e., PolyU Palmprint Database, GPDS Hand Database, and the Bosphorus Hand Database, for a total of 1,072 images. There are 302 textural features found in the palmprint images, and 12 geometrical features are extracted from the hand images. Classification models include Naïve Bayes, Support Vector Machine (SVM), K-Nearest Neighbor (IBk), Decision Tree, Random Tree, Random Forest, and Bagging. The train and test method is used to evaluate the performance of different classifiers. It is observed that Naïve Bayes, SVM, IBk, and Random Tree models result in classification accuracy of 99.44 % with palmprint images using the 302 textural features over the combined dataset. After feature reduction, similar accuracy is achieved with the top ten, and even with the top five, features. For geometrical features, an accuracy of 99.81 % is achieved with the hand images using Naïve Bayes, SVM, IBk, and Random Tree.