Sahil Kailas Shah
@siu.edu.in
Assistant Professor in Computer Engineering
Symbiosis Institute of Geoinformatics, Symbiosis International (Deemed University), Pune
Sahil completed his Bachelor of Engineering in Information Technology in 2010 from Savitribai Phule Pune University with distinction. He completed his Masters in Computer Engineering in 2012 from Savitribai Phule Pune University with distinction. He has been an academic topper throughout his career. Sahil has around 12+ years of teaching experience at undergraduate level. He has also taught various courses at postgraduate level. He likes to learn and share recent technologies in the IT world. He has participated and received certificates of participation in various STTPs, ISTE workshops, Conferences, Seminars and Symposia’s. He has 08 International & 03 National Publications. He is currently associated with Symbiosis International(Deemed University), Pune as an Assistant Professor in Computer Engineering.
EDUCATION
3X Microsoft Certified
B.E.(Information Technology)
M.E. (Computer Engg.)
GATE(CSE) Qualified,
Fellow, IASc Bangalore
RESEARCH, TEACHING, or OTHER INTERESTS
Computer Engineering, Computer Vision and Pattern Recognition, Artificial Intelligence, Information Systems
Scopus Publications
Scholar Citations
Scholar h-index
Scopus Publications
Bhuvan Singh Mangat, Sahil K. Shah, Vidya Kumbhar, and T.P. Singh
IEEE
The main goal of this study is to improve the user experience by providing personalized product recommendations based on user preferences and behaviors. The study examines the various methods and frameworks commonly used in e-commerce recommendation processes, including collaborative filtering, content filtering, and using these methods in neural networks system. Additionally, the proposed algorithm solves the “cold start problem” by newly visited users. Results obtained by proposed approach that use popularity-based algorithm to create the initial set of recommendations highlights how effective the system is at providing accurate and relevant recommendations to users, ultimately increasing customer satisfaction and sales conversion rates of supply. These findings and insights contribute to the development of e-commerce recommendation systems, and present possible avenues for future research and development.
Sahil Kumar Shah, Richa Jain, Vineet Yadav, Aditya Kumar, Pankaj Singh, and Pulkit Tikmani
IEEE
Detecting lung cancer sooner is very much essential for improving the survival rates and reducing the mortality rate as well. This study offers an extensive approach for detecting stages of lung cancer as early as possible from CT scan images using Deep learning. Data set used in this research consists of CT scans of healthy individuals as well as of those diagnosed with various stages of lung cancer. Elastic transformation is used as preprocessing to address class imbalance. Data augmentation is applied to improve model generalization after splitting the data into training, testing and validation. Three pre-trained convolutional neural network architecture (DenseNet201, VGG16, EfficientNetB7) are employed, with transfer learning utilized to fine-tune the models for lung cancer multi-class classification. In the end, weighted CNN ensemble is implemented to combine the predictions of individual models. Popular deep learning metrics such as accuracy, recall, f1-score and precision is used to access the models performance. Final result showcases effectiveness of proposed deep learning model, with the weighted ensemble method achieving an accuracy of 98.19% on the test data. Future work includes exploring different hyper parameter tuning, incorporating additional CNN architectures, and further preprocessing techniques to enhance model accuracy.
Niyaj M. Attar, Sahil K. Shah, Vijayatai Hukare, Vidya Kumbhar, and T. P. Singh
IEEE
India is a leading force in agriculture, responsible for more than 80% of the global crop production. However, it experiences challenges like as droughts and floods that damage the crops. Many farmers cultivate a single crop continuously and apply fertilizers without a clear understanding of its composition. This study explores the application of advanced ML techniques to determine the optimal fertilizers for agricultural use. We evaluated three commonly used techniques: Decision Trees, Random Forest, and Multilinear Regression, to forecast the most appropriate fertilizers: Nitrogen, Phosphorus, and Potassium that should be applied. ML techniques were evaluated for individual fertilizer recommendations (N, P, and K) as well as a combined recommendation approach. It is observed that RF achieved an accuracy of 93.68%, followed by DT at 89%, but MLR underperformed with an accuracy below 50%. This study attempts to assist farmers in increasing their crop yields by providing tailored guidance on fertilizer use. As evident from the obtained results, this study shows the potential to significantly enhance the sustainability of farming and guarantee sufficient food supply on a global scale.
S. K. Shah, V. Kumbhar, and T. P. Singh
International Digital Organization for Scientific Information (IDOSI)
Reetam Ghosh, Sahil K. Shah, and Vidya Kumbhar
IEEE
Liver tumor segmentation(LiTS) is a critical task in medical imaging analysis, playing a key role in the early detection, diagnosis, and treatment planning of liver diseases. Accurate segmentation of liver tumors from medical images is vital for quantifying tumor size, monitoring disease progression, and assessing treatment outcomes. However, manual segmentation is time-consuming and subjective, highlighting the need for automated and reliable deep learning-based methods. This study aims to evaluate and compare the performance of deep learning architectures: U-Net and its variants (U-Net, U-Net++, and newly introduced U-Net Square, and U-Net Cube) for segmenting liver tumors using medical imaging data. Each variant’s unique architectural modifications are examined, emphasizing their potential benefits for liver tumor segmentation tasks. The efficacy and performance of these models for detecting liver tumors is analyzed using metrics like Dice Coefficient. It is observed that basic U-Net outperforms other architectures in liver tumor segmentation. This study further provides potential implications for enhanced diagnostic accuracy and treatment planning in clinical practice. The results contribute valuable insights into the strengths and limitations of U-Net variants, assisting researchers in selecting optimal deep-learning models for liver tumor segmentation.
Mihir Singh, Sahil K. Shah, Vidya Kumbhar, T. P. Singh, and Deepali Tomar
IEEE
With the rising growth in population, the need for food production is increasing day by day. This fosters the need for sustainable food production. However, agricultural products are majorly hampered by diseases on crops. To handle such diseases, automated crop disease identification systems can play a pivotal role. This study attempts to provide a comprehensive exploration of the application of geo-intelligent techniques for automated disease identification. The study presents Northern Leaf blight (NLB) disease recognition in Maize crops using advanced deep-learning techniques applied on high-resolution maize crop leaf imagery obtained through UAVs. Deep learning techniques involving sequential model and transfer learning approaches were explored on high-resolution imagery to classify maize leaves into diseased and healthy categories. It is observed that transfer learning models like Xception performed well in comparison with other models for precise disease identification. Additionally, the study has also explored the potential of YOLOv8 which demonstrates minimal loss with comparable performance. An accuracy of 81.1%, 80.7%, 80.3%, 78.9%, 78.8%, and 75.9% is observed using Xception, CNN, ResNet50, NASNet Mobile, VGG16, and YOLOv8, respectively. This study contributes valuable insights into automated NLB recognition and also holds implications for revolutionizing disease detection across the broader landscape of Indian agriculture. The utilization of advanced deep-learning methodologies offers a promising avenue for safeguarding crop health and enhancing food security in the nation.
Sahil K. Shah, Vidya Kumbhar, and T. P. Singh
IEEE
World Population is increasing drastically and is expected to be around 12.3 billion in 2100. With the increase in population, it becomes important to foster the basic needs of mankind suitably. To withstand larger food requirements, getting higher crop yields is important. However, challenges like crop diseases, environmental factors like soil type, soil moisture, nutrient contents in soil, climate change, etc., affect various crop yields severely. The modern technologies can help in minimizing the adverse effects of these factors on crop yields. The advent of advanced automation techniques in Artificial Intelligence (AI) & Machine Learning (ML) can help in automated detection of crop diseases more accurately. However, use of such automated techniques face several challenges. Current study attempts to present a systematic literature review (SLR) of the existing body of knowledge in crop disease recognition & classification using automated techniques. The review followed PRISMA guidelines & searched four major databases with a comprehensive search query. 60 relevant research studies were identified after filtering through systematic inclusion & exclusion criteria for further analysis. Each research article is examined for the crop species used, the dataset used, & the methodology used. In each of the studied research papers, we looked at the performance measures utilized to assess the overall acquired findings for crop disease diagnosis & classification. It is observed that such systems have great potential to improve food security & agricultural productivity. However, more research is needed to overcome the challenges & limitations of the current methods and to ensure their validity & reliability in different settings and platforms. The readers of this study could acquire challenges faced by researchers in automated crop disease identification systems & their potential solutions to improve the performance of advanced algorithms.
Shubham Shedekar, Sahil K. Shah, and Vidya Kumbhar
IEEE
This research aims to explore the impact of customer reviews on consumer decisions in the e-commerce era. Machine learning techniques, including Naive Bayes, Random Forest, Decision Tree, Extra Trees, and Logistic Regression, are utilized for sentiment analysis of customer reviews. A comprehensive dataset is collected through web scraping and subjected to exploratory data analysis (EDA) to gain a deeper understanding of its characteristics. Feature extraction techniques are applied to convert raw text data into meaningful numerical representations. The ensemble learning approach, specifically the voting ensemble method, is employed to combine individual model predictions, enhancing overall performance and robustness. The findings contribute valuable insights into customer sentiment, empowering businesses to understand preferences and enhance their offerings.
Harsh K. Thakur, Sahil K. Shah, and Vidya Kumbhar
IEEE
With the increasing growth in population, it is important to foster food requirements with ever-growing demand. Food items are perishable and can cause severe issues to mankind. In view of this, such food items need to be analyzed and insights from this analysis can be very important for food production industries. This study attempts to analyze and forecast the non-durable food items in the food industry using advanced machine learning techniques in order to provide valuable insights and practical implications for businesses in various sectors. For analysis and forecasting, a larger dataset comprising the samples from the years 1972 to 2020 is utilized which enabled a comprehensive analysis of long-term trends. Various forecasting models, including simple exponential smoothing, double exponential smoothing, triple exponential smoothing, ARIMA, SARIMA, MLP, LSTM, and CNN, were developed and evaluated using established evaluation parameters. Ensemble learning techniques were also employed to enhance forecast accuracy. The results obtained using the ensemble technique are comparable with existing techniques and contribute to the tailored analysis and development of a forecasting framework for non-durable food item industries. This study can help industries in production planning, inventory management, and meeting consumer demand in order to optimize operations and make informed decisions.
Sampurna Mondal, Sahil K. Shah, and Vidya Kumbhar
IEEE
Peer-to-Peer (P2P) lending has emerged as an attractive investment option for higher returns, contrasting with conventional bank deposits. However, the inherent risk of loan and interest non-repayment necessitates a thorough understanding of credit risk factors. This research delves into the credit risk analysis of one of Europe's prominent P2P lending platforms, “Bondora.” Leveraging publicly available loan application data, supervised machine learning techniques are employed to investigate and identify pivotal factors affecting credit risk. The study evaluates the performance of cutting-edge machine learning models, including Random Forest, Logistic Regression, XGBoost, Regularized Logistic Regression, Decision Tree, and Naïve Bayes, in predicting credit risk within the “Bondora” platform. Notably, XGBoost emerges as the most effective model, boasting the highest accuracy and AUC in loan default prediction. Additionally, Random Forest showcases exceptional recall in detecting default instances. Through this comprehensive analysis, the research facilitates the selection of the most suitable credit risk prediction method for P2P lending networks like “Bondora.” The study's insights empower investors to make informed decisions when considering future investments in such platforms, aiding in the management of potential risks and optimizing investment strategies.
Vijayatai Hukare, Vidya Kumbhar, and Sahil K. Shah
Springer Nature Switzerland
Arbaz Sayed, Vidya Kumbhar, Swapnil Jadhav, and Sahil K. Shah
IEEE
Agriculture Stakeholders need to have a precise understanding of the field boundaries. This is a required resource for farm owners to onboard farms on modern farming software programs, it enhances multiple cropping categorization accuracies and it is used by government agencies to track incentives and food production, along with many other uses. The awareness of the location of fields on the farms is of great importance to a farmer. Even though this can be done for small farm holders by manually surveying the field but it is a very human resource-intensive and time-consuming task for huge farms. Knowing the partition of the fields by their serial number or name can significantly improve the management of fields especially when it is reaping season. Also, the boundary shapefiles can be integrated with a precision farming platform through which farmers can keep a track of all the activities taking place in all their fields, which was not possible before. The current study uses modern state-of-the-art techniques integrated with geospatial technology which helps to reduce the time required for the same process by 95%. The deep learning approach helped to achieve 93% accuracy over the manually drawn fields.
Vyankatesh V. Rampurkar, Sahil K. Shah, Gyankamal J. Chhajed, and Sanjay Kumar Biswash
IEEE
Text information in natural scene images serves as important clues for many image-based applications such as scene understanding, content-based image retrieval, assistive navigation, and automatic geocoding. However, locating text from a complex background with multiple colors is a challenging task. Two different techniques using morphological operations are proposed to find text strings from any natural scene images. Proposed techniques are based on siblings' method i.e. adjacent character grouping method and text line grouping method. Text line grouping method can locate text strings situated at arbitrary orientations. Proposed techniques can find text strings by using structure-based partition and grouping methods using morphological operations. Proposed system strives toward Morphological methodologies that aid automatic detection, segmentation and recognition of visual text entities in complex several images and thus resulting in optimal performance as compared with existing techniques.