Browsing by Author "Trivedi, Sandeep"

Now showing 1 - 8 of 8

A Novel Front Door Security (FDS) Algorithm Using GoogleNet-BiLSTM Hybridization
(IEEE, 2023-02-23) Paula, Luiz Paulo Oliveira; Faruqui, Nuruzzaman; Mahmud, Imran; Whaiduzzaman, Md.; Hawkinson, Eric Charles; Trivedi, Sandeep
Security has always been a significant concern since the dawn of human civilization. That is why we build houses to keep ourselves and our belongings safe. And we do not hesitate to spend a lot on front-door locks and install CCTV cameras to monitor security threats. This paper presents an innovative automatic Front Door Security (FDS) algorithm that uses Human Activity Recognition (HAR) to detect four different security threats at the front door from a real-time video feed with 73.18% accuracy. The activities are recognized using an innovative combination of GoogleNet-BiLSTM hybrid network. This network receives the video feed from the CCTV camera and classifies the activities. The proposed algorithm uses this classification to alert any attempts to break the door by kicking, punching, or hitting. Furthermore, the proposed FDS algorithm is effective in detecting gun violence at the front door, which further strengthens security. This Human Activity Recognition (HAR)-based novel FDS algorithm demonstrates the potential of ensuring better safety with 71.49% precision, 68.2% recall, and an F1-score of 0.65.
A Novel Lightweight Lung Cancer Classifier through Hybridization of DNN and Comparative Feature Optimizer
(IEEE, 2023-05-25) Trivedi, Sandeep; Patel, Nikhil; Faruqui, Nuruzzaman
The likelihood of successful early cancer nodule detection rises from 68% to 82% when a second radiologist aids in diagnosing lung cancer. Lung cancer nodules can be accurately classified by automatic diagnosis methods based on Convolutional Neural Networks (CNNs). However, complex calculations and high processing costs have emerged as significant obstacles to the smooth transfer of technology into commercially available products. This research presents the design, implementation, and evaluation of a unique lightweight deep learning-based hybrid classifier that obtains 97.09% accuracy while using an optimal architecture of four hidden layers and fifteen neurons. This classifier is straightforward, uses a novel self-comparative feature optimizer, and requires minimal computing resources, all of which open the way for creating a marketable solution to aid radiologists in diagnosing lung cancer.
A Novel Sedentary Workforce Scheduling Optimization Algorithm using 2nd Order Polynomial Kernel
(IEEE, 2023-04-05) Patel, Nikhil; Trivedi, Sandeep; Faruqui, Nuruzzaman
No two humans are identical. There are variations in their capability, thinking process, and personality. That is why human society is diverse. This diversity is visible everywhere, including in office environments. The office environments are diverse based on culture, service, and goals. Offices, where the workforce uses computing devices to perform their responsibilities, are mostly sedentary settings where employees must remain seated during office hours. Perseverance and self-motivation are mandatory to make sedentary office hours effective. However, these qualities are not common to everyone. It causes an imbalance in workload distribution and scheduling which facilitates the possibility of irrelevant performance evaluation. This paper addresses these issues and proposes a novel algorithm to optimally schedule the workforce in a sedentary office environment. The proposed algorithm uses 2 nd order polynomial kernel-based Support Vector Machines (SVM) classifier and classifies human activities with 95.0% accuracy. This accurate classification is further utilized to optimally schedule the workforce, which improves performance by 30.6% and saves an average of 1 hour and 2 minutes per day.
An Exploratory Analysis of Effect of Adversarial Machine Learning Attack on IoT-enabled Industrial Control Systems
(IEEE, 2023-05-05) Trivedi, Sandeep; Tran, Tien Anh; Faruqui, Nuruzzaman; Hassan, Md. Maruf
Machine Learning (ML)-based Intrusion Detection Systems (IDS) is an effective technology to automatically detect cyber attacks in the Internet of Things (IoT) dependent Industrial Control Systems (ICS). It is faster, more efficient, and can detect attacks without human intervention. However, ML-based IDSs have introduced another security threat called Adversarial Machine Learning (AML). An AML attack may cause severe industrial infrastructural and production damage resulting in substantial financial loss. This paper presents an exploratory analysis of initiating an AML attack using adversarial samples created using a Fast Gradient Sign Method (FGSM). The research presented in this paper has been conducted from a dataset generated from a full-fledged singular module of a power distribution industry controlled by IoT-enabled ICSs. We explored the AML attack on Gradient Boosting (GB) and Iterative Dichotomiser 3 (ID3) model and discovered the average classification accuracy, precision, recall, and F1-scores are 87%, 88%, 87.5%, and 87%, respectively. The AML attack reduces the average precision, recall, and F1-score by 20.5%, 20.5%, and 22.5%, respectively, when 50% perturbations are added to 10% samples.
An Innovative Deep Neural Network for Stress Classification in Workplace
(IEEE, 2023-04-05) Patel, Nikhil; Trivedi, Sandeep; Faruqui, Nuruzzaman
Human Resource & Management (HRM) plays a vital role in organizational operations. The HRM tries to produce optimal output from human resources through workload balance. One of the core factors of workload balance is stress management. Although Deep Learning technology has introduced revolutionary applications in different sectors, its application in HRM is still nominal. This paper proposes an innovative application of Deep Learning to classify stressed and satisfied employees automatically. This generalized adaptive method utilizes quantitative measures which ensure unbiased classification with 88.40% accuracy and 0.8728 F1-score. The proposed network outperforms similar approaches, paving the path to applying Deep Learning based solutions to ensure a better workplace and proper workload balance through an effortless automatic but reliable stress classifier.
Bacterial Strain Classification using Convolutional Neural Network for Automatic Bacterial Disease Diagnosis
(IEEE, 2023-02-22) Trivedi, Sandeep; Patel, Nikhil; Faruqui, Nuruzzaman
Diseases caused by bacterial contamination are common causes of human illness. Different bacterial strains are responsible for different types of diseases. There are more than 4,900 different strains so far have been discovered. That is why it is impractical to start the treatment of diseases caused by bacterial attacks without diagnosing the particular strain that caused the diseases. The traditional method of bacterial strain classification from the specimens is still widely used in microbiological practice for clinical application. However, it s a time-consuming process and requires well-trained, experienced microbiologists. This paper proposes a computer-aided artificial intelligent-based automatic bacterial strain classification method that is faster than traditional methods and a potentially better alternative. We designed, optimized, and experimented with a Convolutional Neural Network (CNN) to automatically classify bacterial strains from the digital images of the bacterial strains captured using an SC30 camera from an Olympus CX31 Upright Biological Microscope. The proposed network classifies the bacterial strains with 95.12% accuracy, 96.01% precision, 96.70% recall, and 4.88% error rate. This paper uses an innovative image augmentation method to overcome the limitation of the number of training images. The proposed network performs better than similar approaches regarding classification accuracy and network simplicity.
Human Activity Recognition Using Smartphone SensorsHuman Interaction and Classification via K-Ary Tree Hashing Over Body Pose Attributes Using Sports Data
(Springer Nature, 2023-05-25) Trivedi, Sandeep; Patel, Nikhil; Faruqui, Nuruzzaman; Tahir, Sheikh Badar Uddin
Human interaction has always been a critical aspect of social communication. Human action tracking and human behavior recognition are all indicators that assist in investigating human interaction and classification. Several features are considered to analyze human interaction classification in images and videos, including shape, the position of the human body parts, and their environmental effects. This paper approximated different human body key points to track their occurrence under challenging situations. Such tracking of critical body parts requires numerous features. Therefore, we first estimated human pose using key points and 2D human skeleton features to get full human body features. The extracted features are then served to t-DSNE in order to eliminate the redundant features. Finally, the optimized features are infused into the recognizer engine as a k-ary tree hashing algorithm. The experimental results have shown significant results on two benchmark datasets, including the UCF Sports Action dataset with an accuracy of 88.50% and an 89.45% mean recognition rate on the YouTube Action database. The results revealed that the proposed system had achieved better human body part tracking and classification when compared with other state-of-the-art techniques.
NDNN based U-Net
(Daffodil International University, 2023-05-03) Trivedi, Sandeep; Patel, Nikhil; Faruqui, Nuruzzaman
Identifying and segmenting brain tumors using multi-sequence 3D volumetric MRI scans is time-consuming and challenging. Deep learning-based automatic image segmentation approaches are promising solutions to segment brain tumors from MRI 3D reconstructed images. However, T1, T1c, T2, and FLAIR modalities, along with High Graded Gliomas (HGG) and Low Graded Gliomas (LGG), make automatic brain tumor segmentation using deep learning a challenging task. A novel Nested Deep Neural Network (NDNN) has been designed, implemented, and experimented with in this paper, along with an innovative Multimodality Fusion Network (MFS Net). The proposed network segments brain tumors from 3D volumetric images and imposes the extracted feature map on the 3D region with 90.02%, 85.11%, and 85.41% dice score for Whole Tumor (WT), Core Tumor (CT), and Enhancing Tumor (ET) respectively. The novel architecture, innovative multimodality fusion, and outstanding performance of the proposed methodology have been studied, demonstrated, and compared in this paper.