Sandhya Tarwani

@bennett.edu.in

Assistant Professor and School of Computer Science Engineering and Technology
Bennett University

Sandhya Tarwani


            Download Compact PDF  
https://researchid.co/sandy1392

RESEARCH INTERESTS

Machine Learning, Software Engineering, COVID, Maintainability Prediction, Data Science
16

Scopus Publications

Scopus Publications

  • GA-DFS: A Genetic Algorithm-Aided Deep Feature Selection Framework for Chest X-Ray Disease Classification
    Ishaan Garg, Manayu, Ansh Gaur, Sandhya Tarwani
    Lecture Notes in Networks and Systems, 2026
  • Determination of optimum refactoring sequence for maximizing the maintainability of object-oriented systems using machine learning algorithms
    Sandhya Tarwani, Anuradha Chug
    International Journal of System Assurance Engineering and Management, 2025
  • Ensemble Machine Learning Model for Predicting Postpartum Depression Disorder
    Anand, Yash Sharma, Vansh Jain, Sandhya Tarwani
    2024 IEEE Region 10 Symposium Tensymp 2024, 2024
    This study explores the development of an ensemble machine learning (ML) model to predict Postpartum Depression (PPD) disorder, leveraging chi-square test driven feature selection techniques and a diverse array of ML algorithms. Initially, chi-square test is employed to select the most influential features for PPD prediction. From a pool of candidate features, nine key attributes demonstrating the highest association with PPD are identified for model input. Subsequently, eight distinct ML techniques, including K Nearest Neighbors (KNN), Support Vector Machine (SVM), Random Forest (RF), Logistic Regression (LR), Decision Tree (DT), XG Boost (XGB), and Artificial Neural Network (ANN), are applied to develop individual predictive models. Following the training and evaluation of individual models, an ensemble approach is adopted to combine the strengths of multiple algorithms, enhancing prediction accuracy and robustness. The ensemble model aggregates predictions from diverse ML techniques, leveraging their complementary strengths to yield more accurate and reliable predictions of PPD risk. Performance evaluation metrics, such as accuracy, precision, recall, and F1-score, are employed to assess the efficacy of the ensemble model in comparison to individual ML algorithms. The results demonstrate the effectiveness of the ensemble approach in improving prediction accuracy and generalization capability, thereby offering a valuable tool for early identification and intervention in PPD cases. This research contributes to the advancement of predictive modelling in mental health by showcasing the utility of ensemble ML techniques in PPD prediction. The findings underscore the potential of feature selection and ensemble modelling in enhancing the accuracy and effectiveness of PPD risk assessment, thereby facilitating proactive interventions to support maternal mental health.
  • A Comparative Analysis of Machine Learning and Deep Learning Approaches in Deepfake Detection
    Mudit Vashistha, Sarthak Jain, Shubham Pandey, Aryan Pradhan, Sandhya Tarwani
    2024 IEEE Region 10 Symposium Tensymp 2024, 2024
    Deepfakes refer to the visual media where the faces, bodily movements have been digitally altered using some software or program, this has proven to be more of a double edged sword as it also contributes towards content creation and media creation that may be used for positive purposes. To combat this situation, measures to detect deep fake in the media is a credible approach. This work showcases a comparative analysis among 3 Deep Learning as well as 3 Machine Learning algorithms in order to reach a conclusive state of determining the best algorithms that can be implemented for Deepfake detection. For the machine learning algorithms, KNN, SVM and Logistic Regression have been used whereas CNN, TCN and CNN + LSTM have been used for the Deep Learning Algorithm. Detection of deepfakes through these algorithms works by sequentially processing, analyzing and classifying the features on the basis of the dataset fed for the algorithms. The chosen metrics for performing a comparison between each of the algorithms are Accuracy and F1 Score. The development, implementation and comparison of the algorithms was carried out on Google Collab and Jupyter Notebook. Upon comparative analysis of the algorithms between each other, it was found that CNN had the highest accuracy and Fl-score of 0.9409 and 0.7225 respectively with KNN being the worst-performing algorithm with an accuracy 0.5770 and F1 score of 0.4088 respectively.
  • Real-Time Wrong-Way Vehicle Detection System with Automatic Number Plate Recognition for Enhanced Road Safety
    Krish Gaur, Miran Ahmad Siddique, Krishna Beernally, Nityam Madaan, Sandhya Tarwani
    2024 15th International Conference on Computing Communication and Networking Technologies Icccnt 2024, 2024
    In the pursuit of enhancing road safety, the development of real-time detection systems has become imperative. This paper presents a smart approach to wrong-way vehicle detection utilizing computer vision technology for enhanced road safety. The proposed system presents a combination of YOLO for vehicle and license plate detection and OpenCV for license plate recognition, specifically tailored for Indian vehicle number plate dataset. By integrating YOLO’s single-pass object detection capabilities, the system achieves real-time performance, in order to take immediate intervention in traffic violations. Utilizing methods such as centroid tracking and ‘entry-exit’ analysis, the system effectively differentiates between vehicles traveling in correct lanes and those in wrong lanes, enabling immediate enforcement measures. Centroid tracking assigns and tracks centre points of vehicle bounding boxes across frames, ensuring computational efficiency and simplicity in implementation. The ‘entry-exit’ approach utilizes designated lines in video frames to determine the direction of vehicle movement, leading towards accurate wrong-way detection. Moreover, the system leverages EasyOCR for text extraction, ensuring high accuracy in capturing license plate information. This integration of technologies allows for comprehensive monitoring of traffic patterns and swift identification of violations, contributing to proactive enforcement strategies and improved road safety. Through extensive experimentation on the Indian vehicle number plate YOLO annotated dataset, the proposed system demonstrates robust performance in real-world scenarios, exhibiting high accuracy and reliability in wrong-way vehicle detection and automatic number plate recognition. This research represents a significant advancement in leveraging computer vision for proactive traffic management and underscores the potential of technology in enhancing road safety measures.
  • Application of Deep Learning models for Code Smell Prediction
    Sandhya Tarwani, Anuradha Chug
    2022 10th International Conference on Reliability Infocom Technologies and Optimization Trends and Future Directions Icrito 2022, 2022
    Software developer always strives for quality of the software as it tends to be more robust and easier to maintain. Code smells play as a hinder to the quality of the software as they are the surface indication of deeper problem in the source code and hence required to be removed as early as possible. Refactoring is one of the methods to improve the quality of the software without affecting its external behavior. It removes bad smells present in the code as they are surface indication of deeper problem which may lead towards the failure of the software. In this work, we have proposed an approach to predict code smells in the source code using Deep learning algorithms. More specifically, we have trained three deep learning algorithms (CNN, LSTM and MLP) with open-source dataset. Information Gain feature selection algorithm is also applied to get the most prominent attributes for the code smell prediction. After evaluating the performance of these algorithms, results shows that CNN outperforms all other deep learning algorithms with the accuracy of 99% on the selected dataset. This study would be useful for software developer team as they can utilize CNN algorithm in code smell prediction. Results would also help in predicting the code smell which in turn will help in the removal of these smells to enhance the quality of the software.
  • Identifying the Optimal Refactoring Dependencies Using Heuristic Search Algorithms to Maximize Maintainability
    Anuradha Chug, Sandhya Tarwani
    International Journal of Software Engineering and Knowledge Engineering, 2021
    Bad smells represent imperfection in the design of the software system and trigger the urge to refactor the source code. The quality of object-oriented software has always been a major concern for the developer team and refactoring techniques help them to focus on this aspect by transforming the code in a way such that the behavior of the software can be preserved. Rigorous research has been done in this field to improve the quality of the software using various techniques. But, one of the issues still remains unsettled, i.e. the overhead effort to refactor the code in order to yield the maximum maintainability value. In this paper, a quantitative evaluation method has been proposed to improve the maintainability value by identifying the most optimum refactoring dependencies in advance with the help of various meta-heuristic algorithms, including A*, AO*, Hill-Climbing and Greedy approaches. A comparison has been done between the maintainability values of the software used, before and after applying the proposed methodology. The results of this study show that the Greedy algorithm is the most promising algorithm amongst all the algorithms in determining the most optimum refactoring sequence resulting in 18.56% and 9.90% improvements in the maintainability values of jTDS and ArtOfIllusion projects, respectively. Further, this study would be beneficial for the software maintenance team as refactoring sequences will be available beforehand, thereby helping the team in maintaining the software with much ease to enhance the maintainability of the software. The proposed methodology will help the maintenance team to focus on a limited portion of the software due to prioritization of the classes, in turn helping them in completing their work within the budget and time constraints.
  • Illustration and detection of exception handling bad smells
    Sandhya Tarwani, A. Chug
    Proceedings of the 2021 8th International Conference on Computing for Sustainable Global Development Indiacom 2021, 2021
    Code smells signal that something is wrong in the source code that requires immediate attention. Over the years, many research studies have been carried out in this field using various tools and techniques. Exception handling is a powerful mechanism for handling run-time errors, which improves the robustness to enhance software's quality. However, exception handling design problem is the least explored research area that requires sincere intervention. It is often ignored by the developers, thereby having a negative impact on the software concerning reliability, maintainability, etc. In order to understand the exception handling design smells, nine different types of smells, namely careless cleanup, dummy handler, empty catch block, exception thrown in finally block, nested try statement, unprotected main, return code, ignored checked exception and catch clause as spare handler in java have been illustrated to provide a clear view of these smells with the help of the proposed design template. This design template will help software practitioners understand exception handling bad smells more efficiently. The suggested refactoring can further be applied to improve the source code and hence software maintainability.
  • Application of AO* Algorithm in Recognizing the Optimum Refactoring sequence for examining the effect on Maintainability: An empirical study
    Sandhya Tarwani, Anuradha Chug
    Proceedings of the Confluence 2021 11th International Conference on Cloud Computing Data Science and Engineering, 2021
    Bad smells are an indication of deeper problems in source code that need to be identified in order to decrease the accumulation effect in the SDLC which implies that at each stage smells may transform into bugs, faults or even failure of the working software resulting in loss of efforts. Refactoring, on the other hand, helps in removal of smells without affecting the external attributes of the software. Nowadays, researchers are focusing on the detection of optimum refactoring sequences well in advance so that software maintenance cost can be reduced and subsequently the efforts may minimize. In this paper, authors have identified a total of eleven bad smells present in the critically affected class selected on the basis of prioritization technique. After that, an attempt have been made to find optimum refactoring technique sequence using AO* algorithm which will eliminate identified bad smells and thereby helping the team to complete project within budget and time constraints. The obtained results showed that there is a considerable amount of improvement in maintainability value after applying optimum refactoring sequence on every class. This approach will help researchers and practitioners to use heuristic algorithms in finding the sequences in the early phase and hence maintain the source code under surveillance.
  • Determination of optimum refactoring sequence using A∗ algorithm after prioritization of classes
    Anuradha Chug, Sandhya Tarwani
    2017 International Conference on Advances in Computing Communications and Informatics Icacci 2017, 2017
    Bad smells are the surface indication of deeper problem into source code; therefore, they need to be identified as early as possible without compromising on the quality of the software. This lead towards the requirement of refactoring that is the process used in improving the internal attributes like maintainability of the software without affecting its external attributes. Hence, to enhance quality in terms of maintainability refactoring should be done in a controlled and iterative manner. In this study, we have proposed a method that will help researchers and developers to generate a refactoring sequence in advance with the help of heuristic search A∗ algorithm. We have chosen one class of an open source project with the help of prioritization technique to illustrate the generation of the sequence. A∗ algorithm helps in finding an appropriate sequence which has maximum value of maintainability by choosing a path of minimum metrics value. We have identified ten bad smells and used nine refactoring techniques to remove them. With the help of this technique, software developers and maintainers team would be able to figure out refactoring sequence in advance and hence will help them in completing their work within time and budget constraints.
  • Prioritization of code restructuring for severely affected classes under release time constraints
    Sandhya Tarwani, Anuradha Chug
    India International Conference on Information Processing Iicip 2016 Proceedings, 2017
  • Investigating the effectiveness of greedy algorithm on open source software systems for determining refactoring sequence
    Ceur Workshop Proceedings, 2017
  • Predicting maintainability of open source software using Gene Expression Programming and bad smells
    Sandhya Tarwani, Anuradha Chug
    2016 5th International Conference on Reliability Infocom Technologies and Optimization Icrito 2016 Trends and Future Directions, 2016
  • Sequencing of refactoring techniques by Greedy algorithm for maximizing maintainability
    Sandhya Tarwani, Anuradha Chug
    2016 International Conference on Advances in Computing Communications and Informatics Icacci 2016, 2016
  • An empirical investigation of evolutionary algorithm for software maintainability prediction
    Ashu Jain, Sandhya Tarwani, Anuradha Chug
    2016 IEEE Students Conference on Electrical Electronics and Computer Science Sceecs 2016, 2016
  • Agile methodologies in software maintenance: A systematic review
    Informatica Slovenia, 2016