Kapil Sharma
@dtu.ac.in
Professor Department of Information Technology
Delhi Technological University
RESEARCH, TEACHING, or OTHER INTERESTS
Engineering, Artificial Intelligence, Computer Engineering, Multidisciplinary
Scopus Publications
Scopus Publications
Anurag Chauhan, Kapil Sharma, and Sudhanshu Choudhary
Elsevier BV
Rahul Gupta, Kapil Sharma, and Ramesh Kumar Garg
MDPI AG
The widespread integration of smartphones into modern society has profoundly impacted various aspects of our lives, revolutionizing communication, work, entertainment, and access to information. Among the diverse range of smartphones available, those operating on the Android platform dominate the market as the most widely adopted type. With a commanding 70% share in the global mobile operating systems market, the Android OS has played a pivotal role in the surge of malware attacks targeting the Android ecosystem in recent years. This underscores the pressing need for innovative methods to detect Android malware. In this context, our study pioneers the application of rough set theory in Android malware detection. Adopting rough set theory offers distinct advantages, including its ability to effectively select attributes and handle qualitative and quantitative features. We utilize permissions, API calls, system commands, and opcodes in conjunction with rough set theory concepts to facilitate the identification of Android malware. By leveraging a Discernibility Matrix, we assign ranks to these diverse features and subsequently calculate their reducts–streamlined subsets of attributes that enhance overall detection effectiveness while minimizing complexity. Our approach encompasses deploying various Machine Learning (ML) algorithms, such as Support Vector Machines (SVM), K-Nearest Neighbor, Random Forest, and Logistic Regression, for malware detection. The results of our experiments demonstrate an impressive overall accuracy of 97%, surpassing numerous state-of-the-art detection techniques proposed in existing literature.
Soumya Bajpai, Kapil Sharma, and Brijesh Kumar Chaurasia
IEEE
The Internet of Things (IoT) is a distributed network of physical entities ("things") for exchanging data over the Internet. These things may be equipped with sensors, actuators, software, and communication technologies. The IoT network can deliver substantial benefits to our society and economy, including better health services and safer societies. However, some security and safety challenges are rapidly increasing, and their solutions are not clearly defined. The work uses genetic algorithms to address intrusion detection systems (IDS) for IoT networks. It is a hybrid combination of a genetic algorithm and K-Nearest Neighbors (KNN) technique to solve the IDS in IoT networks with the aim of lightweight and accurate performance. The proposed algorithm has been analyzed and compared with the other existing models. Extensive simulation results demonstrate our suggested algorithm’s effectiveness and ability to attain 99% accuracy.
Akshay Mool, Jeebananda Panda, and Kapil Sharma
World Scientific Pub Co Pte Ltd
Modern face detection algorithms fail to provide optimal results when they have to deal with larger amounts of data per frame while processing higher quality videos. This paper tackles that problem and offers a solution to deploy commercially used state-of-the-art face detection algorithms to process only the regions of interest in a frame, and discard the rest to decrease the data to be processed. The model maintains the accuracy of the base algorithm while decreasing the processing time per frame, thereby increasing the overall efficiency. The selection of region of interest is dependent on the detection of facial window in the previous frame. Therefore, the choice of base algorithm plays an important role in determining the speed of the framework. The model achieves increased processing speeds of about 69–76% more than the standalone usage of the detection algorithms for analyzed frame rates.
Ritcha Saxena, Priya Verma, Vikas Sharma, Kapil Sharma, and Jiya
IEEE
One of the common diseases that many individuals experience is diabetes. One way to think of diabetes is as a collection of physiological a condition where the body's blood glucose level is higher than the standard guidelines for treatment. Typical method to identify the blood samples are drawn to measure blood glucose levels, which is highly painful. The creation of a blood glucose prediction model is the solution for preventing diabetes and may regulate insulin levels. Here, we suggest a novel deep learning forecasting model for the precise diabetes classification by blood glucose level prediction. The proposed model makes use of long short-term memory (LSTM) in identifying both short- and long-term dependencies, as well as convolutional layers to extract meaningful knowledge and learn the internal representation of time-series data. In this research work proposed CNN-LSTM hybrid model's performance is tested against the most advanced deep learning and machine learning models currently in use. According to the preliminary experimental research, using LSTM layers in conjunction with extra convolutional layers could significantly improve forecasting ability.
Daksh Chaudhary, Kapil Sharma, Vikas Sharma, and Prasuk Jain
IEEE
Creating descriptive phrases with visual and textual data is known as image captioning. Transformers use an encoder and decoder configuration to manage language comprehension and machine translation. As part of our effort to create a small and lightweight model that can be deployed with ease, we present the Lightweight Transformer with an embedded GRU decoder for image captioning. We reduce the usual architecture in this model by reducing the number of encoders and decoders to only one encoder and a GRU-integrated decoder. Furthermore, including multilevel rich visual features from Inception V3 enhances the encoder's performance. We conducted a number of thorough experiments to assess the effectiveness of this suggested Lightweight Transformer architecture using the Viz Wiz Captions dataset.
Nilesh Agarwal, Kapil Sharma, and Mukhtiar Singh
IEEE
Biomass fuels offer a viable answer as a sustainable energy source that can significantly decrease reliance on fossil fuels in the future. Biomass higher heating value (HHV) is a key indicator that defines its energy content and suitability for various applications. A diverse range of biomass fuels is included in this dataset identified by their corresponding HHV value and the amounts of hydrogen, nitrogen, carbon, oxygen as well as sulfur. Various deep learning model was trained and tested based on various biomass samples obtained from [1]. The best performing model was found by using several parameters like Mean Squared Error (MSE), Root Mean Squared Error (RMSE) & Mean Absolute Error (MAE). The results showed that the Hyper tuned ANN model had a good accuracy and performance on test data with a low RMSE, MAE and MSE of 0.8828, 0.6524 and 0.7793 respectively. The application of artificial neural networks and deep learning in biomass energy research holds enormous chances for understanding of this vital renewable energy source and optimizing its use for a more sustainable future.
A.V.S. Swetha, Manju Bala, and Kapil Sharma
IEEE
Breast diagnosis from pathology reports is a widely employed clinical method for diagnosing breast tumors. But it's challenging due to low contrast, high noise, and diverse appearances. Experienced professionals rely on factors like global context, local geometry, and intensity changes, acquired through years of clinical experience. To ease the burden on doctors, we propose integrating machine learning with diagnosis practice. However, effective and efficient breast tumor detection is crucial for automated disease diagnosis. In the last decade, various deep learning models have emerged for breast tumor detection. Traditional methods lack real-time responsiveness, accuracy, and scalability. To address these issues, we introduce an enhanced framework with hybrid model based on a multi-task cascaded convolutional network. This framework leverages a substantial dataset of clinically confirmed images with precise labels, employing a multi-task cascaded architecture. It involves two stages of deep convolutional networks designed to detect and recognize tumor affected area progressively, ensuring real-time operation and high accuracy. Additionally, we introduce an improved feature fusion-based augmentation method, enhancing diagnostic model performance. Our experiments demonstrate exceptional accuracy and time efficiency. Furthermore, the model's versatility is highlighted in the context of related diseases. Our proposed model showed a high accuracy of 97.69 for multi classification and 99.39 for binary classification. Also, our proposed framework when compared with many other hybrid model utilized less time (3.19 seconds per image) and resources making it efficient.
Vikas Sharma, Kapil Sharma, and Akshi Kumar
IEEE
In this age of digital and smart healthcare, cutting-edge technologies are being used to improve operations, patient well-being, life expectancy, and healthcare costs. Digital Twins (DT) have the potential to significantly change these new technologies. DTs could revolutionise digital healthcare delivery with extraordinary creativity. A digital representation of a physical asset that is always its digital twin due to real-time data processing. This paper proposes and builds a DT-based intelligent healthcare system that is aware of its environment. This approach is a great advance for digital healthcare and could improve service delivery. Our most notable contribution is a machine learning-based electrocardiogram (ECG) classifier model for cardiac diagnostics and early problem detection. Our cardiac models predict some situations with exceptional accuracy when applied to different ways. These findings highlight the potential for Digital Twins in healthcare to create intelligent, comprehensive, and scalable Health-Systems that improve patient-physician communication. Our ECG classifier also sets a precedent for using Artificial Intelligence (AI) and Machine Learning (ML) to continually monitor wide range of human body data and identify outliers. ECG data processing has improved significantly using neural network-based algorithms over classic machine learning methods. In conclusion, our work integrates digital twins with cutting-edge AI and machine learning to revolutionise healthcare. Future healthcare will be predictive and improve lives.
Rahul Gupta, Kapil Sharma, and R. K. Garg
Tech Science Press
A Divya Sree and Kapil Sharma
Bentham Science Publishers Ltd.
Aims and Background: The topology and communication links of vehicular adhoc networks, or VANETs, are always changing due to the transient nature of automobiles. VANETs are a subset of MANETs that have applications in the transportation sector, specifically in Intelligent Transportation Systems (ITS). Routing in these networks is challenging due to frequent link detachments, rapid topological changes, and high vehicle mobility. Methods: As a result, there are many obstacles and constraints in the way of creating an effective routing protocol that satisfies latency restrictions with minimal overhead. Malicious vehicle detection is also a crucial role in VANETs. Unmanned-Aerial-Vehicles(UAVs) can be useful for overcoming these constraints. This study examines the utilize of UAVs operating in an adhoc form and cooperating via cars VANETs to aid in the routing and detection of hostile vehicles. VANET is a routing protocol. The proposed UAV-assisted routing protocol (VRU) incorporates two separate protocols for routing data: (1) a protocol called VRU_vu for delivering data packets amid vehicles with the assist of UAVs, and (2) a protocol called VRU_u for routing data packets amid UAVs. Results: To estimate the efficacy of VRU routing objects in a metropolitan setting, we run the NS-2.35 simulator under Linux Ubuntu 12.04. Vehicle and UAV motions can also be generated with the help of the mobility generator VanetMobiSim and the mobility simulation software MobiSim. Conclusion: According to the results of the performance analysis, the VRU-protocol is able to outperform the other evaluated routing protocols in terms of packet-delivery-ratio (by 17 percent) &detection-ratio (9 percent). The VRU protocol cuts overhead near 41% and reduces end-to-enddelay in mean of 15%.
Parminder Pal Singh Bedi, Manju Bala, and Kapil Sharma
Springer Nature Singapore
Ganesh P. Prajapat, Vikas Sharma, D. K. Yadav, Surender Singh Tanwar, and K. G. Sharma
Springer Nature Singapore
Aditi Sharma, Kapil Sharma, and Akshi Kumar
Springer Science and Business Media LLC
Anurag Chauhan, Kapil Sharma, and Sudhanshu Choudhary
Springer Science and Business Media LLC
Kapil Sharma and Sandeep Tayal
Springer Science and Business Media LLC
A.V.S. Swetha, Manju Bala, Kapil Sharma, and Rahul Katarya
IEEE
Breast cancer is the most common cancer among women worldwide, underlining the significance of early identification for successful treatment. Deep learning (DL) has shown promising results in breast cancer prediction, but traditional DL models struggle with imbalanced datasets making the model biased. To overcome this problems hybrid models are built but, these hybrid models often consume huge resources making them computationally inefficient. This study introduces an innovative breast cancer classification framework using deep convolutional neural networks. The framework relies on weight factors and threshold values to create an effective hybrid model. Initially, two separate deep convolutional neural network models are employed, and their test accuracies are compared to a predefined threshold. If both models accuracy falls below the threshold, a hybrid model is constructed. This hybrid model merges features from both models through a multimodal fusion approach, expanding the feature set but potentially affecting computational efficiency. To address this efficiency challenge, an optimal feature selection algorithm is employed to choose the most relevant features from the expanded set. Empirical evidence validates the framework’s excellence, even when dealing with imbalanced datasets, as it surpasses evaluation criteria. The suggested hybrid model achieves an impressive binary classification accuracy of 99.69% while maintaining a minimal processing time of just 3.52 seconds.
parminder Pal Singh Bedi, Manju Bala, and Kapil Sharma
IEEE
Background: The most often type of cancer detected in women is Breast cancer. It is formed in the breast cells. It is an acutely life-threatening disease in women as compared to other cancers. Cancer caused in the breast is grouped into Various categorizations of breast cancer as per their cell's appearance (i) invasive ductal carcinoma (IDC) and (ii) ductal carcinoma in situ (DCIS), $2^{\\mathrm{n}\\mathrm{d}}$ one being generally having no negative effects and is formed quite slowly, whereas the 1st one, IDC type is comparatively more dangerous and surrounds the breast tissues to great extent. Around 80% of the breast cancer patients fall under this category [1]. Scope: In the context of medical data, where comprehensive information about breast cancer and its symptoms is available, the scope of our research becomes especially relevant. This field requires efficient methods for the detection of the disease to support healthcare professionals in delivering effective diagnoses. Problem: The fundamental challenge lies in the need for timely and precise patient disease detection summaries during the diagnostic process. This problem has been a longstanding concern in the healthcare industry. The core issue at hand is the inefficiency and resource burden posed by the current manual methods of summarizing medical data during diagnosis. These methods are neither time-effective nor cost-efficient, often leading to delays in patient care and a drain on valuable resources. Overall Contribution: Our proposed method aims to Improve the validation score with loss function for breast cancer detection. This innovation will significantly benefit the research community and healthcare experts by streamlining the process of diagnosis. By saving time and resources, our contribution promises to enhance the overall efficiency of healthcare practices, leading to more effective and timely patient care.
A.V.S. Yeswanth, Himanshu Nandanwar, Kapil Sharma, and Anamika Chauhan
IEEE
Rising air pollution and increased concentrations of airborne particulate matter pose a significant threat to both developing and developed countries, adversely affecting air quality and human health. In response to this pressing issue, we have developed a cost-effective Air Quality Monitoring System (AQMS) prototype with a focus on low power consumption. The prototype utilizes real-time data monitoring through the Thingspeak platform. Employing various data-driven machine learning techniques such as Principal Component Analysis (PCA), Technique for Order of Preference by Similarity to Ideal Solution (TOPSIS), and Canonical Correlation Analysis (CCA), the system evaluates the collected data. PCA is employed to calculate principal components, while TOPSIS assesses positive and negative distances. This comprehensive approach enables a thorough analysis of air quality, contributing to a better understanding of the environmental impact on human well-being.
Srideep Das and Kapil Sharma
IEEE
Skin cancer is a common and potentially life-threatening disease that requires early detection for successful treatment. Recently, various techniques involving artificial intelligence (AI) and Deep learning (DL) have been applied to skin cancer detection, offering the potential to enhance the performance and speed of diagnosis. In this paper we discuss different forms of skin cancer and their detection challenges and also presented a comprehensive review of the literature on various AI and DL techniques used for detection. We proposed a novel AI-based approach called Inception-ResNet-V2 for early skin cancer detection. The proposed approach is a variant of convolutional neural network (CNN) architecture to analyze skin lesion images and accurately identify the presence of cancerous cells. Using a publicly accessible dataset, we assessed the performance of the suggested strategy and contrasted it with currently available state-of-the-art techniques. Our findings show that the suggested strategy performs the best and is superior to existing techniques in all the areas. Overall, this research paper highlights the potential of AI and DL techniques for skin cancer detection and proposes a promising new approach that can contribute in enhancing the performance and effectiveness of skin cancer diagnosis. The suggested method may be applied in clinical settings and can assist dermatologists in making quicker and more precise diagnosis of skin cancer.
Sachin Kumar, T. Senthil Siva Subramanian, Kapil Sharma, and G.V. Ramesh Babu
CRC Press
Vikas Sharma, Kapil Sharma, and Akshi Kumar
IEEE
Yashasvi Chahal, Ritika Tokas, and Kapil Sharma
IEEE
Advancement in technologies like big data, Internet of Things (IoT) and cloud computing has led to the Digital Twin (DT) being prominently used from idea to practice in numerous industrial fields as a precision simulation model. Substantial simulation models have been suggested for Digital Twin in numerous fields in the past 12 years. In the domain of IoT and healthcare, researchers are keen to find smart automated IoT-based solutions for Healthcare systems. So far, numerous works highlight the problem of DT in healthcare systems through theoretical frameworks. However, application-based solutions and validations are hardly provided. In this paper, we propose the Digital Twin in smart healthcare systems for Diabetic Retinopathy and a framework for the same to enhance healthcare procedures and improve the health of patients. For the first time, we developed a Diabetic Retinopathy identifier and classifier Digital Twin to detect and diagnose damaged retina due to diabetes with the help of Deep Learning and IoT. We used EfficientNet for Transfer Learning leading to a good accuracy score being predicted. The results accumulated towards the end have proven that fusion of healthcare and Digital Twin might result in better processing of healthcare tactics by putting patients and healthcare experts in conjunction in a rational, extensive, and expandable medical ecosphere. Additionally, executing this Digital Twin offers the foundation for making the use of DL, IoT and AI with various human shape metrics for ceaseless tracking, anomaly detection as well as ideating treatments.
Shresth Goyal, Jiya Sharma, Kapil Sharma, and Akshi Kumar
IEEE
Remote healthcare monitoring has become a reality due to advancements in sensors and computation power. A person can track their own, or their family member’s health through mobile applications as well. Physical healthcare monitoring depends upon machines and medicines and tests, whereas most of the psychological disorders can’t be detected through these tests, so we need to monitor a person’s day to day behavior for accurate analysis. Their exist some mobile applications that ask a user to answer a few questions daily regarding their day-to day activities and emotions they were feeling while performing those tasks. Mobile applications for psychological health tracking have become increasingly popular in recent years. These apps can be used to monitor and track various aspects of mental health, such as mood, anxiety, stress, sleep, and physical activity, providing users with valuable insights into their mental health and wellbeing. UI/UX design plays a crucial role in the success of any mobile app, including psychological health monitoring apps. Emotion recognition, mood recognition, stress and anxiety levels predicted by these apps rely on user input to accurately fill the questionnaire based upon their feelings. An intuitive and engaging user interface can increase user engagement and improve the accuracy of psychological health tracking.
Anamika Chauhan and Kapil Sharma
Springer Nature Singapore