This implementation presents a game recommendation system that utilizes natural language processing (NLP) techniques to provide personalized game suggestions based on user preferences. The system processes a dataset of games containing descriptions and emotional tones to determine relevant recommendations. It employs TF-IDF (Term Frequency-Inverse Document Frequency) vectorization to transform textual data into numerical representations, enabling meaningful comparisons between game content and user input. The cosine similarity metric is then used to assess the closeness of games to the given preferences. The recommendation process begins by filtering the dataset based on the user's specified emotional tone, ensuring that only games matching the desired sentiment are considered. Next, the input content is vectorized, and similarity scores are computed against all games in the dataset. The system ranks the games based on similarity and retrieves the top five matches. This approach enhances game discovery by allowing users to find games that align with both their emotional state and gameplay preferences, offering a more personalized gaming experience.
The enquiry develops an effective sleuthing system for "Autism Spectrum dieses(ASD)" during early level habituate Machine Learning methodologies despite challenge in full removal of the diseases while attempting early interposition for stiffness reduction. The system includes four "Max Abscaler,PowerTransformer,QuantileTransformer,NormalizerandFeature Scaling(FS)” that analyze their effects on “four standard ASD datasets” from diverse “age gatherings (Babies, Teenagers, Youngsters, and grownup). ” "The study algorithmic rule (Ada Boost, K-Nearest neighbor, Decision Tree, Random Forest, Gaussian Naïve Bayes, support vector machine, Logistic Regression, Linear Discriminant Analysis)" are use on let in descale datasets. The organization results are assessed employ numerous actual mensuration, distinguishing which classifiers and part purpose operation work substantially with each age grouping of participant. The testing results show striking precision ascent, with the “Voting classifier anticipating the ASD nearly elevated accuracy of sister and toddler”, while it accomplishes the most elevated accuracy for teenagers and adult. The challenge involves acomprehensive examination of the grandness of each element employ four component determination strategies. It suggests that the constituent research can serve hospital therapy experts in the production of conclusion-making through ASD viewing and suggests the implication of raise ML technique for the prediction of ASD across various geezer hood group and socioeconomic backgrounds. The urge structure exhibits ideal consequence in direct contrast with current techniques for other ASD screening. The aim scheme should work on the heartiness and accuracy of ASD recognition; a troupe technique “Adaboost and Random Forest with voting classifier" was perform, execute an uncommon “100% accuracy”.
We introduce, in this work, a holistic approach for music recommendation by multi-modal emotion recognition. With facial expressions, speech, and text sentiment analysis as the three inputs, our system dynamically matches the music recommendation with the emotional state of the user. This system ensures accuracy with real-time performance due to its use of the latest machine learning techniques, including CNNs, LSTMs, and pre-trained NLP models. Its effectiveness is well supported through experiments that compare it with conventional methods and provide a personalized and emotionally aware user experience.
Recommendation systems are a form of data filtering that plays a crucial role in helping users discover relevant content,based on their interests and needs. They have become an integral feature of numerous platforms, particularly in digital streaming services, where users often struggle with decision fatigue due to vast content. However, challenges such as the coldstart problem, data sparsity, and evolving user preferences require advanced approaches that balance accuracy and scalability.This study presents a hybrid movie recommendation system that integrates content-based filtering and collaborative filtering techniques to generate highly relevant suggestions. The content-based approach leverages metadata from the Movie_Metadata and TMDB datasets, incorporating movie popularity and other relevant features. Collaborative filtering analyzes user interactions and preferences from the Movie_Metadata ratings dataset. By combining these techniques, the system effectively mitigates individual limitations, resulting in a robust and efficient recommendation engine. Experimental results demonstrate that the hybrid model outperforms standalone methods in user satisfaction. This scalable solution addresses the growing demands of streaming platforms and highlights the broader potential of hybrid recommendation systems across various domains.
The goal of this project is to provide a real-time sign language translator so that people who use sign language may interact with others. Leveraging advanced machine learning techniques, the system recognizes fingerspelling hand gestures, converts them into text, and subsequently synthesizes the text into speech. MediaPipe is employed for real-time hand tracking and gesture recognition, while a Convolutional Neural Network (CNN) classifies the gestures. The recognized text is converted into speech using a Text-to-Speech (TTS) library, facilitating both visual and auditory communication. This system promotes inclusivity by enabling seamless interaction for speech-impaired individuals, fostering better integration into daily life activities.
Lung diseases continue to pose a major global health issue, necessitating prompt and precise diagnosis to enhance patient outcomes. This project introduces a deep learning approach for the automated detection and classification of lung diseases, utilizing the Inception-v3 and ResNet50 architectures. A model uses convolutional neural networks (CNNs) to analyse chest X-rays and identify lung conditions like COVID-19, Tuberculosis, Pneumonia, and healthy cases. The dataset is subjected to comprehensive preprocessing and augmentation to ensure its robustness, and the transfer learning capabilities of both architectures improve performance even with limited data. The model is trained and optimized using categorical cross-entropy loss along with the Adam optimizer, achieving an accuracy of 94.92% with Inception-v3 and 92.99% with ResNet50 on the testing dataset. This research underscores the promise of deep learning in enhancing medical imaging diagnostics, providing a scalable, efficient, and dependable solution for predicting lung diseases.
As a leading cause of cancer-related deaths worldwide, lung cancer requires prompt and precise diagnosis to improve survival rates. In this extended study, we enhance lung cancer detection by incorporating advanced deep learning models for both classification and localization. The Xception model is employed for classification, achieving an impressive 99% accuracy in identifying cancerous and non-cancerous lung tissues. To further improve tumor detection, YoloV5 and YoloV8 are integrated, enabling precise localization of affected regions in CT scan images. This combined approach ensures a robust and comprehensive solution for lung cancer diagnosis. A user-friendly Flask framework with SQLite integration is also created, allowing users to seamlessly interact with the system through a secure signup and login mechanism. This facilitates real-time testing and efficient image processing, making the solution practical for medical applications. By leveraging these advanced models and an intuitive interface, the proposed system significantly enhances the accuracy, reliability, and usability of lung cancer detection, providing a valuable tool for early diagnosis and treatment planning.
In recommendation systems, the Cold Start issue is a major obstacle, especially when working with new users or objects that have no interaction history. This study tackles the problem by applying a Transfer Learning methodology, employing pre-trained models like BERT to extract rich semantic elements from movie-related textual data, such as actors, directors, and overviews. The study improves the system's capacity to produce precise recommendations even in the lack of a significant user-item interaction history by utilizing these cutting-edge natural language processing techniques. As part of our methodology, we perform intensive data preparation on a large IMDb dataset that includes movies from 1951 to 2023. The extensive movie metadata and wide variety of genres in the dataset offer a strong basis for model evaluation and training. By including BERT for feature extraction, followed by similarity measurements utilizing cosine similarity and PCA for dimensionality reduction, the suggested system shows enhanced performance in addressing the Cold Start phenomenon. The results indicate that recommendation systems' accuracy and adaptability are greatly increased by transfer learning, providing a scalable solution for dynamic, data-poor contexts.
Vitamin deficiency is a terrible disease that kills people all over the world because it causes aberrant cells to develop and spread out of control. An essential organ that creates a barrier of defence against the environment is the skin tissue. However, the skin tissue is vulnerable to illness since it is found on the exterior. The most deadly type of vitamin deficiency in people is vitamin deficiency. If caught early, stage 1 vitamin deficiency can be totally treated. Which one is cancerous can only be determined by a skilled dermatologist. Additionally, which one is not cancerous? The development of new moles or modifications to preexisting moles are typical signs of Stage 1 vitamin deficiency. Performing a physical examination with dermoscopy is one of the initial procedures in diagnosing Stage 1 Vitamin Deficiency. Without a dermatoscope, it is exceedingly difficult to visually identify these Stage 1 vitamin deficiencies since their boundaries are frequently blurry. A variety of pre-processing and image filtering techniques are used to the dermoscopy picture of vitamin deficiency. Segmentation is used to distinguish the area impacted by the vitamin deficiency from the healthy skin tissue. When it comes to helping medical professionals diagnose and treat patients, medical photographs are essential. Techniques for digital image processing can more precisely detect the characteristics and offer the relevant illness status.
One important area of research is deepfake audio detection, which separates real human voices from speech that has been modified or produced artificially. Generative models like WaveNet, Voice Conversion, and Text-to-Speech (TTS) synthesis have greatly enhanced the quality and realism of deepfake audio due to the quick development of artificial intelligence. This has raised significant ethical and security issues in a number of domains,including media, cybersecurity, and forensic investigations. In order to analyze Mel-Spectrograms and MelFrequency Cepstral Coefficients (MFCCs), this study suggests a deepfake audio detection framework that makes use of Convolutional Neural Networks (CNN) and Bidirectional Long Short-Term Memory (BiLSTM) networks. These attributes enable the model to accurately discriminate between synthetic and real audio by capturing the spectral and temporal aspects of speech. A balanced training environment is ensured by the 94,734 audio samples in the dataset utilized in this study, which is evenly distributed between actual and false recordings. To improve model performance, preprocessing methods such time-frequency domain analysis, feature scaling, and noise removal are used. Using demanding experimental settings, the suggested CNN-BiLSTM architecture is trained and assessed, obtaining 98% accuracy and proving its resilience in identifying deepfake speech. In order to prevent audio forgeries and improve the security of voice-based authentication systems, the results of this study emphasise the significance of hybrid deep learning architectures. In order to increase the scalability and versatility of deepfake detection models, future research will investigate the merging of self-supervised learning strategies with real-time detection methods.
The integrity of information on social media platforms is seriously threatened by deepfake technology, which manipulates media using artificial intelligence. Deepfake content has proliferated in India, particularly in the political and entertainment sectors, where disinformation has been spread by AIgenerated movies and fake news. The main goal is to create a strong artificial intelligence model that can reliably identify deepfake content on social media sites, with an emphasis on machine-generated tweets that use FastText embeddings. Conventional techniques included manual social media post filtering using preset criteria and keyword matching, human moderation, and fact-checking organisations. These techniques lacked the scalability to handle the enormous volume of online content and were timeconsuming and frequently wrong. Deepfake and AIgenerated content detection done by hand is incredibly ineffective, prone to mistakes, and unable to process the enormous amount of social media data in real time. As a result, false and damaging information may proliferate before being discovered or eliminated. The goal of this study is to counteract false information and protect the integrity of online debate in light of social media's increasing power to shape public opinion. By automating the study of social media information, deep learning models in particular can greatly enhance the detection of deepfakes. While deep learning models can be used to determine if a tweet was created by an AI or a human, FastText embeddings will transform tweets into meaningful word vectors. This approach enables real-time detection, greater accuracy, and scalability compared to existing approaches
Monkeypox is an emerging zoonotic disease that has raised global health concerns due to its increasing transmission rates. Traditional diagnostic methods rely on laboratory testing, which can be timeconsuming and inaccessible in resource-limited settings. This study presents an Optimized Deep Neural Framework (ODNF) to diagnose monkeypox based on clinical symptoms, leveraging deep learning for accurate and rapid classification. The research explores various machine learning models, including Random Forest, XG Boost, and Cat Boost, before implementing ODNF, which achieved superior performance with a 99% accuracy rate. The dataset underwent preprocessing steps, including handling imbalanced data and feature encoding, ensuring optimal learning. Additionally, Local Interpretable Model-Agnostic Explanations (LIME) was employed to enhance model interpretability, providing insights into symptom-based predictions. Comparative evaluation against traditional models demonstrated that ODNF outperforms existing approaches, making it a viable AI-based diagnostic tool for monkeypox detection.
Road traffic accidents in India remain a significant public health issue, resulting in frequent injuries and fatalities. Quick intervention is critical for improving survival rates, yet delays in emergency responses often obstruct timely care. This idea proposes an automated system that leverages computer vision to analyze real-time CCTV video streams, enabling fast accident detection and response. Our approach involves selecting key frames from the video, extracting important features, and using machine learning models to classify frames as containing accidents or not and as suggested in the future scope of the base paper we focus on integrating an automated alert dispatch system, which will trigger notifications to emergency services upon accident detection. This alert mechanism will provide real-time geolocation data of the incident ensuring rapid response and communication.
The rapid advancements in generative AI have led to the development of dedicated models for content, image, music, and video creation. However, customers are often faced with difficulties in switching between devices to meet multi-modal content generation. ZumbleBot bridges this gap by combining content, image, music, and video creation into one, integrated platform. Using cutting-edge Huggingface Pre-trained AI models like Qwen for content, Steady Dissemination for images, MusicGen for music, and text-to-video models, ZumbleBot uncouples creative workflows and enhances openness. The platform constitutes a literary insight and creates returns over unique groups of media while ensuring proper coherence. This article analyzes the engineering, demonstrate integration, and application of ZumbleBot, as well as its uses in content creation, education, and advertising. Also, we examine the challenge of multimodal AI age and suggest arrangements to maximize execution and maintain yield quality. ZumbleBot addresses a step toward steady, expert, and astutely AI-powered imagination. With the use of cutting-edge generative AI, ZumbleBot redefines multi-modal creativity, making content generation with AI more accessible and efficient.
The increasing use of medical imaging for diagnosing brain tumors has highlighted the need for effective and privacy-preserving machine learning models. This study presents a solution for brain tumor classification from MRI scans using federated learning, which ensures patient data privacy by keeping the data decentralized and only sharing model updates. A deep learning approach, utilizing transfer learning with the VGG-16 architecture, is employed to classify images into different tumor types: pituitary, meningioma, glioma, and no tumor. Data augmentation techniques are applied to enhance the training dataset, addressing class imbalance issues. The model is trained in a federated learning environment with multiple clients, where each client trains the model on their local dataset, and the results are aggregated to update the global model. After training, the model's performance is evaluated using metrics such as classification accuracy, precision, recall, F1-score, and a confusion matrix. This approach allows for high accuracy in tumor classification while ensuring data privacy through federated learning. The proposed method aims to facilitate efficient, privacy-preserving medical image classification in a distributed setting, particularly in scenarios with sensitive data.
The increasing reliance on digital transactions has led to a surge in credit card fraud, which poses significant financial risks to individuals and institutions. Traditional fraud-detection systems often struggle because of the highly imbalanced nature of transaction datasets, in which fraudulent cases represent only a small fraction of all transactions. This study addresses these challenges by developing a machine-learning-based fraud detection system using Logistic Regression and adaptive training strategies. To improve the detection accuracy, the Synthetic Minority Oversampling Technique (SMOTE) was applied to handle class imbalance by generating synthetic fraudulent samples. Additionally, feature scaling using StandardScaler ensures that all features are normalized for better model performance. The proposed system was trained and evaluated using various performance metrics, including the accuracy (96.49%), precision (98%), recall (95%), and F1-score (96%). The model achieved an impressive ROC-AUC score of 0.9935, demonstrating its ability to effectively distinguish between legitimate and fraudulent transactions. The results highlight the effectiveness of combining Logistic Regression with data preprocessing techniques to enhance fraud detection in imbalanced datasets.
Traditional phishing detection methods struggle to keep pace with evolving cyber threats. An LSTM-based deep learning system in order to identify fraudulent websites are presented in this study. The model effectively captures temporal dependencies and contextual patterns within webpage content, distinguishing phishing websites from legitimate ones. Additionally, an attention mechanism enhances feature extraction, improving detection accuracy. In comparison to more traditional methods, our results show significant improvements in accuracy, precision, and recall across a variety of datasets. The model's adaptability to emerging phishing strategies through continuous learning makes it a robust solution for phishing detection.
In today's globalized world, the demand for precise, adaptable, and scalable translation tools is at an all-time high. This project presents an advanced multilingual translation platform optimized for domain-specific and multimodal tasks, harnessing cutting-edge Natural Language Processing (NLP) techniques and the transformative capabilities of Marian MT—a state-of-the-art transformer-based architecture renowned for its efficiency, scalability, and contextual precision.The system supports diverse input formats, including text-to-text, imageto-text, and audio-to-text translations, making it an indispensable solution for specialized domains such as healthcare, law, and academia. To enhance non-textual input processing, the platform incorporates Convolutional Neural Networks (CNNs) for precise feature extraction and superior contextual understanding.By leveraging fine-tuned domain-specific datasets and a feedback-driven continuous improvement mechanism, the system delivers unmatched translation accuracy, adaptability, and scalability. Rigorous evaluations using BLEU, ROUGE, and other performance metrics confirm its superior accuracy and contextual fidelity across diverse languages and input modalities.Designed for inclusivity and user-friendliness, the platform serves a wide range of stakeholders, including individuals, businesses, and organizations. It enhances accessibility, supports global collaboration, and sets a new standard for multilingual translation systems in specialized applications, pushing the boundaries of crosscultural and multimodal communication
This study investigates the performance of a renewable energy-based distributed generation system utilizing an Adaptive Neuro-Fuzzy Inference System (ANFIS) tuned Unified Power Quality Conditioner (UPQC). The system integrates photovoltaic (PV) and wind energy sources to address power quality issues and enhance the reliability of power distribution networks. The ANFIS-tuned UPQC is designed to dynamically optimize its control strategies, responding effectively to fluctuations in power generation and varying load demands. The main goals of this study are to determine how well the system compensates for reactive power loss and maintains voltage regulation, as well as how well it can reduce common power quality issues including harmonics, voltage sags, and swells. The research shows how well the ANFIS-tuned UPQC performs in providing a reliable and high-quality power supply through extensive simulation and analysis. Key findings indicate that the integration of ANFIS with UPQC not only enhances the power quality but also improves the overall efficiency and resilience of the distributed generation system. The system effectively utilizes renewable energy sources, reducing dependency on conventional power generation and contributing to environmental sustainability. The results highlight the potential of this advanced control strategy in promoting the widespread adoption of distributed generation systems based on renewable energy, ensuring both reliability and sustainability in modern power grids
The project is a Flask-based application designed for face recognition and attendance tracking. It includes functionality for training a convolutional neural network (CNN) model using the InceptionV3 architecture, which is fine-tuned to classify faces. The application supports registering users by capturing and saving their facial images, which are then organized into directories. These images undergo data augmentation to enhance model robustness during training. A pre-trained CNN model is employed to detect and identify faces, allowing attendance to be marked automatically based on face recognition results. Attendance data is recorded in an Excel file, where new entries are added dynamically for each date.The application also provides an analysis feature that processes attendance data, generates visual summaries for different dates, and displays trends in a userfriendly format. Interactive web pages enable users to train the model, register faces, upload images for marking attendance, and review attendance statistics. The project incorporates image preprocessing, realtime face detection, and the use of cascaded classifiers, with seamless integration of attendance tracking into the system. The intuitive interface and automated processes aim to enhance efficiency and accuracy in attendance management.
