Authors - Sandeep Shinde, Parth Kedari, Atharva Khaire, Shaunak Karvir, Omkar Kumbhar Abstract - With the use of cutting-edge technologies like Flask, web technology, API rendering, Make It Talk, and machine learning (ML), an AI smart tutor bot is being implemented with the goal of giving users an engaging and customized learning experience. The bot uses machine learning techniques to analyze responses and generates quiz-style questions with multiple-choice possibilities and extended answers. This allows for quick feedback. Additionally, it has an interview mode in which the user engages with an AI avatar that reads their body language and facial emotions. Using written material and specialized alphabets, the AI avatar is dynamically educated, gaining comprehensive knowledge and an accurate evaluation of user performance. The research article goes into detail about the system architecture, how different technologies were integrated, and the process for training the avatar and gauging user response. Through user feedback and experimental trials, the AI Smart Tutor Bot's performance is assessed, showcasing its potential as an advanced teaching tool that can adapt to each student's unique learning needs while boosting comprehension and engagement.
Authors - Bhagyashree D. Lambture, Madhavi A. Pradhan Abstract - Phytochemical qualities, geographic information, environmental conditions, and traditional medicinal knowledge are some of the sources of information that are incorporated into this research project, which presents a comparative examination of machine learning (ML) algorithms for the qualitative evaluation of medicinal plants. In order to categorize and forecast the medicinal value of plants based on multi-modal data, the purpose of this study is to investigate the effectiveness of various machine learning algorithms. For the purpose of determining which method is the most effective for evaluating complicated and diverse datasets, a full evaluation is carried out utilizing well-known machine learning models. These models include decision trees, random forests, support vector machines, and deep learning algorithms. Key criteria including as accuracy, precision, recall, F1-score, and computing efficiency are utilized in order to evaluate the levels of performance achieved by each method. For the purpose of gaining a deeper comprehension of the role that each data source plays in determining the medicinal potential of plants, the value of features and their interpretability are also investigated. A basis for the ongoing development of AI-driven tools in pharmacological research and plant-based drug discovery is provided by the findings of this comparative analysis, which offer vital insights into the usefulness of machine learning for medicinal plant assessment. Contributing to the expanding fields of computational botany and natural product science, the purpose of this study is to improve the precision and effectiveness of the evaluation of medicinal plants.
Authors - Azhar Abbas, Farha Abstract - Fraudulent claims in the insurance industry lead, to significant financial losses and negatively affect both policyholders and insurance firms. Machine learning has proven to be revolutionizing fraud detection since it is more than just performing the ordinary rule-based systems while automating and optimizing detection processes. The current work proposes a novel hybrid approach that combines supervised and unsupervised techniques in machine learning with applications in accurately and robustly detecting insurance fraud. Three primary models include in the framework are Decision Tree, Random Forest, and Voting Classifier, which improve detection performance on real-world datasets. In addition, an embedding-based model interprets sequential claims data, and a statistically validated network is used to detect patterns of collusion and fraud among related entities. Extensive experimentation was conducted using large-scale motor, and general insurance datasets and showing that the proposed hybrid model achieved an accuracy of 89.60%. Hyperparameter tuning and data preprocessing were used to further refine the model's performance; it was able to counterbalance all issues brought forth by imbalances, complexities, and complexities due to variations in fraud types. The methodology outperformed the existing models, better at identifying rare sophisticated cases of fraud. The practical implications of deploying machine learning models in the insurance sector are also discussed from the angle of best practices for data governance, model interpretability, and stakeholder trust. In Future this work will be improved by incorporating real-time analytics to provide quicker detection, enhancing interpretability features, and adapting the model to emerging fraud patterns in evolving data environments.
Authors - Poonam Yadav, Meenu Vijarania, Meenakshi Malik, Neha Chhabra, Ganesh Kumar Dixit Abstract - Parkinson's disease is aging-associated degenerative brain illness that results in the degeneration of certain brain regions. Early medical diagnosis of Parkinson's disease (PD) is difficult for medical professionals to make with precision. Magnetic Resonance Imaging (MRI) and single-photon emission computed tomography, or SPECT are two medical imaging strategies that can be used to non-invasively and safely assess quantitative aspects of brain health. Strong machine learning and deep learning methods, along with the efficiency of medical imaging techniques for evaluating neurological wellness, are necessary to accurate the identification of Parkinson's disease (PD). In this study, we have used dataset of MRI images. This study suggests three deep learning models: ResNet 50, MobileNetV2 and InceptionV3 for early diagnosis of PD utilizing MRI database. From these three models, MobileNetV2 demonstrated superior accuracy in training, testing and validation with a rate of 99%, 94%, and 96%, respectively. With its effectiveness and precision, MobileNet V2 has a lot of potential for PD identification using MRI scans in the future. We may further advance the development of dependable and easily accessible AI-powered solutions for early diagnosis and better patient care by tackling the issues and investigating the above-mentioned future paths.
Authors - Aye Thiri Nyunt, Nishi Vora, Devanshi Vaghela, Brij Kotak, Ravi Chauhan, Kirtirajsinh Zala Abstract - This paper is an AI and Machine Learning Algorithm - based dualistic Gesture-to-Speech and Speech-to-Gesture framework. The core of this initiative is to enable machines and humans to converse with each other by enabling the translation of physical body movements into reasonable speech and vice versa. We used deep learning models- Convolutional Neural Networks (CNN)- to train our system using a dataset consisting of human gestural movements and the relevant speech patterns. For the Gesture-to-Speech module, real-time gesture recognition and interpretation were used, which involved computer vision and were implemented to interpret gestures into speech output containing words and phrases representing the message illustrated by the gestures. The Speech-to-Gesture module, on the other hand, uses speech as input to produce context-related gestures-the main purpose of which is to improve user interaction and experiences. In the system, multiple applications were tested, including sign language and webcams. Further research will try to extend the flexibility of the system to include various languages, cultural backgrounds and characteristics of individual gesture styles which eventually has a high level of customization. We had designed the CNN architecture for real-time gesture recognition and taken care of data preprocessing as well to increase accuracy concerning different types of gestures. We created Gesture-to-Speech translation with the use of an LSTM, then added in a Text-to-Speech engine for it to have a very natural sound. We then worked on Speech-to-Gesture and even refined the gestures through a CNN-based network, to ensure transitions are very fluid. Everything was coordinated such that there would be synchronous gestures and speech for extremely natural real-time interaction. We coached on how one would integrate, test, and further optimize models with dropout and batch normalization for higher performance.
Authors - Varun Maniappan, Praghaadeesh R, Bharathi Mohan G, Prasanna Kumar R Abstract - This paper constitutes a comprehensive review of how language models have changed, focusing specifically on the trends toward smaller and more efficient models rather than large, resource-hungry ones. We discuss technological progress in the direction of language models applied to attention mechanisms, positional embeddings, and architectural enhancements. The bottleneck of LLMs has been their high computational requirements, and this has kept them from becoming more widely used tools. In this paper, we outline how some very recent innovations, notably Flash Attention and small language models (SLMs), addressed these limitations by paying special attention to the Mamba architecture that uses state-space models. Moreover, we describe the emerging trend of open-source language models, reviewing major technology companies efforts such as Microsoft’s Phi and Google’s Gemma series. We trace here the evolution from early models of transformers to the current open-source implementations and report on future work to be done in making AI more accessible and efficient. Our analysis shows how such advances democratize AI technology while maintaining high performance standards.
Authors - S.K. Manjula Shree, Shreya Vytla, J. Santharoopan, Harisudha Kuresan, A.Anilet Bala, D.Vijayalakshmi Abstract - The goal is to use a Random Forest classifier to categorize future price movements as "up" or "down" in order to forecast stock market trends. In order to guide investing strategies, this model will examine pertinent attributes and previous stock data. The effectiveness of Logistic Regression, Support Vector Machines (SVM), and Random Forest Classifier in forecasting stock market movements is compared in this study. The ensemble approach Random Forest is very resilient under erratic market situations since it is excellent at handling noisy, complex data and capturing non-linear patterns. SVM performs best on smaller, more structured datasets, however noise and non-linearity might be problematic. Despite its simplicity and interpretability, logistic regression is constrained by its linear character and finds it difficult to account for the dynamic, non-linear behavior of stock prices. In recall focused tasks, logistic regression is helpful because it performs well in identifying true positives (such preventing missed opportunities in stock predictions). SVM's reliance on kernel functions makes it computationally expensive, but it can also be helpful when handling smaller datasets with clear patterns and where accuracy is needed. All things considered, Random Forest offers the greatest results around 99% especially for difficult stock market prediction assignments.
Authors - Umakant Singh, Ankur Khare Abstract - This paper aims to find the optimal model for stock price forecast. In examining the different approaches and aspects that need to be considered, it is exposed the methods decision tree and Gradient Boosted Trees Model. This paper aims to propose a more practical approach for making more accurate stock predictions. The dataset including the stock bazaar values from the prior year has been considered first. The dataset was optimized for actual analysis through pre-processing. Therefore, the preprocessing of the raw dataset will also be the main emphasis of this work. Again, decision trees and gradient tree models are used on the pre-processed data set and the results thus obtained are analyzed. In addition, forecasting papers also address issues related to the usage of forecasting systems in actual situations and the correctness of certain normal value. This paper also presents a machine learning model for predicting stock stability in financial markets. Successful stock price forecasting greatly benefits stock market organizations and provides real solution to problems faced by investors.
Authors - Aneesh Kaleru, Chaitanya Neredumalli, Mrudul Reddy, Ramakrishna Kolikipogu Abstract - One major risk factor that contributes to traffic accidents globally is poor visibility in foggy situations. Drivers are seriously threatened by fog because it weakens contrast, hides important objects, and makes lane markings almost invisible. Recent developments in visibility enhancement methods for foggy circumstances are summarized in this paper, with a focus on picture defogging combined with object detection and lane aid. We analyze the application of models such as Conditional Generative Adversarial Networks (cGANs), Single Shot Multibox Detectors (SSD), All-in-One Defogging Network (AODNet), and You Only Look Once (YOLO) from the perspective of deep learning and computer vision. These methods have the potential to increase driver safety in inclement weather by identifying impediments, improving visibility, and offering lane guidance. The review also covers the limitations of these solutions, such as computational demands and requirements for real-time processing. Our goal is to provide researchers and practitioners with a comprehensive understanding of the current methods and their uses, allowing for the development of effective visibility enhancement systems that can prevent accidents and save lives.
Authors - Sindhu C, Taruni Mamidipaka, Yoga Sreedhar Reddy Kakanuru, Summia Parveen, Saradha S Abstract - India is a country with very rich ancient historical legacy. It preserved vast cultural and linguistic knowledge through stone inscriptions. Extracting text from ancient stone inscriptions and translating it into a language which is understandable by everyone is a very challenging task due to script variations, natural wear, and the uneven degraded surfaces of stone carvings. Our idea is to build a model which can extract the text from these stone inscriptions which were written in Telugu language and translate them into other Indian local languages. The Region-Based Convolutional Neural Network (R-CNN) model which is integrated with Tesseract OCR is trained on a custom dataset of 30,000 labelled images of Telugu script, encompassing Achulu (vowels), Hallulu (consonants), and Vathulu. By achieving a 96% accuracy in character detection, this model demonstrates significant reliability in recognizing Telugu characters from degraded and complex inscriptions. Data augmentation techniques, including rotations, flips, and shifts were used to further enhance the model’s robustness to different orientations and environmental conditions encountered in historical artifacts. The text which is extracted from the image is ultimately translated into Indian local languages using an API-based translation module, enabling a seamless interpretation of ancient content. This research contributes a comprehensive and automated solution for cultural preservation, giving us a scalable method to digitize and make historical inscriptions accessible to everyone which are integral to Telugu heritage and linguistic history.
