Rice Leaf Disease Early Detection Image Classification Approach Based on Convolutional Neural Network Inderjeet Singh, Yadwinder Singh, Manik Pathria, Udit Kathuria, Rajendra Dhaka, Sourav Thakur Conference Proceedngs Wccst 2026 World Conference on Computational Science and Technology, 2026 Finding diseases in rice leaves on time is very important for getting better crop production. Normally, farmers check the leaves by looking at them, but this takes a lot of time and may not always be correct. It is also difficult to check large fields in this way. With the help of deep learning, we can now create systems that can detect diseases automatically and save time. In this work, we used a CNN model to identify rice leaf diseases using 5,932 images. The images belong to four disease types: Bacterial Blight, Blast, Brown Spot, and Tungro. We created a simple CNN model and trained it from the beginning. We made all the pictures <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$128 \times 128$</tex> so that the model could easily and quickly work with them. The model was trained using standard methods and then tested on data that was not used for training. It got about 96.4% of accuracy, which is a good score. The model's simplicity and lightweight in design. This study shows that a basic Convolutional Neural Network (CNN) can help identify plant diseases, and it has potential for future improvements.
A Novel Random Forest-SMOTE Framework With Polynomial Feature Engineering for Early Detection of Gastrointestinal Disorders Parvez Rahi, Mohammad Rashid Hussain, Salem Alqahtani, Mohammad Husain, Ajay Pal Singh, Inderjeet Singh IEEE Access, 2025 About 40% of people around the world have gastrointestinal (GI) problems. Gastrointestinal issues negatively impact the lives of millions of people and place significant stress on healthcare systems. The wrong diagnosis might make treatment less effective because symptoms like diarrhea, bloating, stomach pain, and liver problems often happen at the same duration. For early intervention and personalized care, it is essential to correctly and quickly classify these symptoms. We used the Random Forest model to create a strong classification framework for this study. It can tell if someone has one of seven main symptom groups: bloating, diarrhea or constipation, blood in the stool, unexplained weight loss, liver problems, nausea or vomiting, or abdominal cramps or pain. In our model we have used a multi-modal dataset that includes genetic markers, clinical assessments, lifestyle factors, and diagnostic imaging data to give a full picture of the diagnosis, and then we added polynomial feature engineering and other methods to the model to make it better at finding hidden and small patterns and relations in the parameters. We have also used the Synthetic Minority Oversampling Technique (SMOTE) to fix the problem of uneven data and GridSearchCV to make the model robust against hyperparameters. Although past studies have mostly focused on the lesion-based imaging or targeted disease identification, our model is a step forward in terms of symptom-based early classification with multi-modal tabular data. The model is capable of combining clinically informed two-factor interactions (e.g., stress × appetite loss, inflammation × bowel habits), which significantly improve the predictive performance and interpretability. This design reduces the disparity between the algorithmic predictions and the clinician thinking, thus, the system has become more applicable to realistic diagnostic decision-making. In this way, it allows for personalized treatment plans that make the disease curable and save time and risk of severity of GI disease. Future work will focus on adding biomarker data and real-time patient tracking to the framework so that it can be used easily in clinical decision-making.
Real-Time Helmet Violation Detection for Traffic Management Using EfficientNet v2 and PerspectiveNet Jaswinder Singh Bhatia, Inderjeet Singh, Yash Mahajan, Devansh Saxena 2025 International Conference on Engineering Innovations and Technologies Icoeit 2025, 2025 Motorcycle accidents significantly contribute to traffic-related fatalities globally, often resulting from noncompliance with helmet regulations. Effective automated detection of helmet violations is crucial to enhance road safety and enforce traffic management efficiently. In this paper, we propose a real-time helmet violation detection system leveraging advanced deep learning techniques. We introduce an optimized PerspectiveNet model integrated with EfficientNet v2 as the backbone, providing a robust and computationally efficient detection architecture. The proposed model addresses challenges in current detection systems, including occlusion, variable lighting conditions, and computational complexity. By adapting PerspectiveNet's density-aware architecture, our approach intelligently distinguishes between high-density images, with complex occlusions, and low-density straightforward scenarios. EfficientNet v2 backbone further reduces computational load while preserving high feature extraction capability. Comprehensive experiments were performed on a carefully annotated subset of the India Driving Dataset (IDD), focusing explicitly on helmet and non-helmet classifications. The experimental setup evaluated multiple optimization strategies, including varying batch sizes, activation functions, and network optimizers. Our results demonstrate exceptional performance, achieving an accuracy of 95.2%, surpassing existing state-of-the-art benchmarks. Furthermore, extensive ablation studies confirm that adopting EfficientNet v2 significantly reduces computational complexity, facilitating real-time deployment even on embedded devices. This study highlights the potential of combining EfficientNet v2 and PerspectiveNet to accurately and rapidly detect helmet violations, significantly contributing toward improved traffic law compliance, reduced accident rates, and enhanced safety measures. Future work includes validation in diverse real-world scenarios and deployment on embedded systems for practical traffic management applications.
"alzheimer's and AI Transforming Diagnostics with Regularized Neural Models" Amit Walia, Sandeep Singh Kang, Parvez Rahi, Inderjeet Singh 2025 4th Opju International Technology Conference on Smart Computing for Innovation and Advancement in Industry 5 0 Otcon 2025, 2025 The World Health Organization estimates that Alzheimer's disease affects over 55 million people globally and that around 10 million new cases are identified annually, making it a global health emergency. The crucial necessity for an early and precise diagnosis to enhance patient outcomes and quality of life is highlighted by this startling incidence. Unfortunately, such conventional methods have taken much time and are considered subjective, therefore, delaying necessary interventions. The current challenge required developing a DNN model able to classify a patient into any of the five major categories of: No Cognitive Impairment (NCI), Mild Cognitive Impairment (MCI), Early-Stage Alzheimer's Disease, Advanced Alzheimer's Disease, or Alzheimer's Mimic or Mixed Dementia. The model applies cutting-edge techniques, such as L2 regularization to prevent overfitting, batch normalization to ensure stable learning, and dropout layers to enhance its ability to generalize. ReLU activation functions in the hidden layers allow the model to identify complex patterns in clinical, cognitive, and behavioural data. After training on a diverse dataset, the DNN achieved an impressive 91% classification accuracy. This work shows the potential of deep learning in revolutionizing the diagnosis of Alzheimer's, providing health care professionals with a reliable tool for personalized care. Future work will include integration of additional data sources, such as imaging and longitudinal records, to further improve the model's accuracy and adaptability in real-world clinical settings.
Time Series Analysis in Functional Genomics Yash Mahajan, Inderjeet Singh, Muskan Sharma, Shweta Sharma Multimodal Data Fusion for Bioinformatics Artificial Intelligence, 2025 Time series analysis has emerged as a vital analytical framework in the field of functional genomics, offering insights into the temporal dynamics of gene expression, protein synthesis, and other molecular interactions. By focusing on the temporal dimension, time series analysis enables researchers to observe changes in biological systems over time, capturing the complex interplay between genes, their products, and external stimuli. This method is especially useful for understanding dynamic regulatory functions, such as patterns in gene expression and protein activity, providing a high-resolution perspective on biological processes. With the integration of multimodal data, such as transcriptomics, proteomics, and metabolomics, time series analysis can further deepen our understanding of cellular mechanisms, offering a holistic view of genetic functions in various physiological and pathological contexts. This chapter aims to explore the foundational concepts, methodologies, and applications of time series analysis in functional genomics, emphasizing the potential of this approach to uncover temporal trends and interactions that are crucial for advancing genomic research. Additionally, the chapter highlights the challenges associated with analyzing dynamic biological data and the benefits of combining different data sources to improve the accuracy and depth of genomic insights.
Dynamic Pricing and Energy Optimization Strategies Inderjeet Singh, Muskan Sharma, Suvigya Yadav, Yash Mahajan, Koushik Sundar Secure Energy Optimization Leveraging Internet of Things and Artificial Intelligence for Enhanced Efficiency, 2025 This chapter delves into the evolving dynamics of energy pricing and optimization strategies that are pivotal for sustainable development in the energy industry. It explores the concept of dynamic pricing, which allows electricity prices to fluctuate based on real-time demand and supply factors, encouraging more efficient energy usage. Various dynamic pricing models, such as time-of-use (TOU), real-time pricing (RTP), and critical peak pricing (CPP), are analyzed for their effectiveness in managing energy demand and integrating renewable energy sources. The chapter also highlights energy optimization strategies, including demand response programs and energy storage systems, which are critical in enhancing grid stability, reducing energy costs, and promoting environmental sustainability. Through the combination of these approaches, the chapter underscores the potential for achieving a more reliable, cost-effective, and sustainable energy future.
Towards Accurate Skin Cancer Screening: A CNN Model for Dermatological Images Inderjeet Singh, Parvez Rahi, Diya Sharma, Yashita Garg, Yash Narayan Proceedings of 2025 International Conference on Computing for Sustainability and Intelligent Future Comp Sif 2025, 2025 Early detection is vital for effective treatment and increased survival rates in the case of skin cancer, which is a common type of cancer globally. Machine learning (ML) has become a promising tool in the field of dermatology, providing improved precision and effectiveness in the screening of skin cancer. This assessment examines the different classification methods used in machine learning for detecting skin cancer, discussing the techniques, outcomes, and obstacles linked to these approaches. The research investigates widely utilized algorithms like CNNs, SVMs, and ensemble methods, analyzing how they are utilized in distinguishing between cancerous and non-cancerous growths. It also touches upon the significance of image preprocessing, feature extraction, and data augmentation in enhancing classification performance. In addition, the analysis discusses present constraints, such as a lack of data, prejudice, and the necessity for transparency in ML models, emphasizing topics for further study. This paper aims to lead future advancements in ML-based skin cancer screening by offering a comprehensive summary of the current state of the field, resulting in more precise, available, and scalable diagnostic options.
An Intelligent Predictive Model for Cyber Physical Systems: A Scalable Framework Using Internet of Medical Technology Inderjeet Singh, Jaswinder Singh Bhatia, Amit Walia, Harkirat Singh, Ajay Pal Singh, Eakansh 2025 8th International Conference on Circuit Power and Computing Technologies Iccpct 2025, 2025 A new intelligent and resilient predictive model for cyber-physical-social systems (CPS) has been designed for ehealth applications. Autonomous services in contemporary healthcare greatly depend on CPS, which couples physical processes with computational systems and human interactions. The Internet of Medical Things (IoMT) has greatly improved CPS-based healthcare by enhancing patient data management and facilitating medical decision-making. Still, some issues like limited power supply, workload uneven distribution, and privacy are the issues that hold back system advancement. Such constraints affect scalability, performance security, and operational overall efficiency. The solution to all these challenges lies in using the proposed model’s edge intelligence feature to improve e-health applications’ performance. Energy consumption is minimized, and data privacy is assured while increasing connectivity, analysis capacities, and forecast service functionalities. The solution presents a reliable connection framework that provides secure data transfer and reinforces digital communication. The solution also includes a low-power algorithm, incorporates artificial intelligence, and implements advanced load-balancing strategies for secure and efficient operation in healthcare information and medical technology systems. Simulation results demonstrate that the model performs better compared to current systems, with better packet reception, lower network overhead, lower latency, and increased reliability. This work represents a significant step toward creating sustainable, scalable, and secure e-health systems, towards smarter and more interconnected healthcare environments.
Text Processing and Analysis Pipeline for Scientific Literature Inderjeet Singh, Satyam, Ashutosh Semwal, Shivansh Singh, Gouri Gupta Proceedings 3rd International Conference on Advances in Computing Communication and Applied Informatics Accai 2024, 2024
