This study presents a multimodal deep learning framework designed to analyze sentiment patterns in YouTube videos and explore their association with early indicators of mental well-being. The approach integrates textual transcripts, vocal characteristics, and facial expressions into a unified representation to capture the emotional depth that individual modalities often miss. The model was trained and evaluated on a curated dataset of diverse YouTube content, and the results consistently showed that the fused architecture performed better than unimodal baselines. Compared with text-only or audio-only systems, the multimodal model achieved higher accuracy and fewer misclassifications, particularly in cases where speakers displayed subtle or mixed emotions. The integration of vocal cues such as pitch variation, speaking rate, and stress patterns helped clarify emotional ambiguity, while visual features such as micro-expressions, gaze direction, and facial tension added further clarity to the sentiment shifts within the videos. Transformer-based fusion delivered the most stable performance, demonstrating strong generalization across varied communication styles and recording conditions. In addition to reporting classification outcomes, the study examined how specific multimodal patterns correlate with non-clinical markers of mental health. Consistent associations were observed between fluctuating sentiment trajectories and indicators such as emotional instability, sustained negative tone, and reduced expressive variability. Instances where facial expressions contradicted verbal sentiment also showed relevance for identifying mild distress signals. These findings suggest that multimodal emotional cues can offer valuable insights into the affective state of content creators and may support research on digital well-being. The analysis also revealed challenges related to background noise, varying video quality, and inconsistent facial visibility, which influenced the reliability of certain features. Despite these limitations, the study demonstrates that combining audio, visual, and textual information provides a more complete and reliable picture of sentiment expression on social media platforms. The proposed framework offers a foundation for future systems aimed at understanding online emotional behavior and contributes to ongoing discussions on the responsible use of machine learning in mental-health-oriented applications.

This research paper presents a comprehensive study that compares different machine-learning models used for detecting credit card fraud. The focus of the study is on the application of federated learning to safeguard user privacy. The models examined in the research include RNN, CNN, LSTM, MLP, RBFNN, GRU, and a novel model called FedMLPNN (Federated MultiLayer-Perceptron Neural Network). The analysis results indicate that the FedMLPNN model outperformed the other models in various key metrics, such as accuracy, precision, F1 score, recall, and AUC. The FedMLPNN model achieved an impressive accuracy rate of 97.34

Anomaly detection is an essential part in medical field, the reason being it is best to identify a disease at its early stages so it can be cured or can be prevented from being worse. But sometimes it could be just a simple variation and not an actual disease, so it is important to distinguish between a disease and a variation. In medical field it is difficult to analyze high dimensional healthcare data along with finding the illness, ailment, etc. To overcome all these problems this paper focuses on a novel approach of using Generative Adversarial Networks (GAN) which can help improve the scope of anomaly detection in medical field. Here the GAN is trained on pre-existing datasets which helps in producing realistic images of the scan. To detect illnesses in their early stages, track patient health trends and avoid invalid diagnostic reports, this study aims to focus on a GAN-based anomaly detection method for a variety of medical data. To make the model reliable that can revolutionize medical diagnostics all issues related to training stability, data diversity and hyperparameters optimization are addressed.

Brain tumour besides being lethal can also affect other human organs on a long-term basis if not detected at an early stage. Based on the region of its presence and speed of growth, it can be classified as Glioma, Meningioma. or Pituitary. In the given work, a Stacked Neural Net approach for the classification of brain tumour is presented. The images are first pre-processed using Histogram equalization and One Hot Encoding. Thereby, the improved images are fed into 3 popular pre-trained deep networks viz. Res-Net-50, VGG-16, and Xception as level-0 models in parallel. The features extracted are then fed into a neural network as level 1 learner for the final classification. A publicly available Figshare dataset has been used for testing the performance. The suggested approach has been thoroughly evaluated against its base models using accuracy, precision, and recall metrics and also the performance has been analysed using accuracy and loss curves along with epochs. The suggested model achieved a validation and training accuracy of 97.3

In the modern digital world, social media has become essential for interpersonal interaction by promoting the interchange of ideas and points of view. But there are difficulties in this digital environment, especially concerning rude behavior and offensive remarks. To address both problems at once, the research focuses on sentiment analysis and abusive comment detection in social media interactions. The dataset contains Hate Speech and Offensive Content Identification (HASOC) data from 2019 to 2021 to identify hate speech in Hindi on various social media platforms. To categorize comments into abusive and non-abusive groups, several BERT models, including mBERT, DistilBERT, RoBERTa, HateBERT, and IndicBERT, have been utilized. Additionally, a comprehensive sentiment analysis of the derogatory comments has been performed. The research presents a stacked ensemble framework for binary (abusive and non-abusive) and multiclass (hate, offensive, and profane) classification that integrates predictions from mBERT, HateBERT, and IndicBERT models. Further, the study provides an integrated approach for providing abusive comment detection and sentiment analysis using a multi-output model. The proposed ensemble model achieves 94% accuracy in binary classification, with precision, recall, and F1 scores all approaching 94%. Multiclass ensemble models yield an accuracy of 93% and associated precision, recall, and F1 scores of 91%, 92%, and 92%, respectively. A comparative analysis using several state-of-the-art techniques has been generated to verify the efficacy of the suggested methodology.

Environmental Sound Classification (ESC) finds a vital application in wildlife conservation, audio-video systems, music instrument classification, automatic speech recognition systems, sound event detection etc. Many conventional model training methods depending upon an enormous amount of annotated data, have been proposed in the literature for the same. The proposed technique focuses on the use of CNN for classifying short audio clips of environmental sounds. Existing models use auditory features like either Log-Mel spectrogram (LM) or Mel Frequency Cepstral Coefficient (MFCC) etc. for the classification or improvement, we have stacked all the different features visualized using the Librosa libraffiry such that it combines all the feature information into one image. The accuracy of the network is evaluated on the ESC-50 dataset of environmental and urban recordings. The stacked features seemed to perform better for the dataset chosen. Three stacked features are provided as input in form of channels to a transfer learned model, which outperforms the CNN models that trained from scratch. The highest precision and recall are obtained for Log-Mel Scale Spectrogram, Spectral Contrast and chroma features which is 95.91% and 95.81% respectively.

The ability to detect cardiac disease early is vital to saving lives. Heart attacks are one of the leading reasons for high death rates worldwide due to the high cost of identifying cardiac disorders, which is crucial to the healthcare industry, as well as the shortage of human and logistical resources. In order to find the best machine learning algorithm for early-stage heart disease prediction, a model is presented in this study. In order to train and evaluate the model based on the data for efficient decision-making, machine learning techniques are utilized. The goal of this research is to combine Principal Component Analysis and Multilevel Perceptron model to propose an updated diagnosis and classification model based on machine learning (ML) for predicting heart failure and heart disease detection. Using cutting-edge techniques and foundational classifiers like Support Vector machines, Random forests, and K-Nearest Neighbors, an empirical evaluation of the suggested methodology was carried out with accuracy, precision, recall, and F1-score serving as the evaluation criteria. The proposed method has shown the maximum accuracy of 95.1

Skin cancer poses a significant burden on mankind and healthcare systems globally, necessitating the development of effective diagnostic and treatment strategies. This paper introduces FusionEXNet, an innovative and interpretable fused deep-learning model for skin cancer detection. FusionEXNet leverages the strengths of EfficientNetV2S and XceptionNet architectures to extract robust features from dermoscopic images, achieving superior performance compared to individual models. While XceptionNet and EfficientNetV2S attained accuracies of 88.82% and 88.01%, respectively, FusionEXNet surpassed them with an impressive accuracy of 90.83%. To enhance model interpretability, Explainable Artificial Intelligence (XAI) techniques, such as SmoothGrad and Faster Score-CAM, are integrated, providing valuable insights into the decision-making process. The model was trained and evaluated using the extensive HAM10000 dataset, consisting of over 10,000 high-resolution images across seven skin lesion categories. The proposed FusionEXNet model offers a reliable, accurate, and interpretable system for skin cancer detection, contributing to improved patient outcomes and more efficient diagnostic processes. This study underscores the potential of combining advanced CNN architectures with XAI methods to create powerful tools for dermatological diagnostics.

Background: Colorectal cancer (CRC) is the second leading cause of cancer fatalities and the third most common human disease. Identifying molecular subgroups of CRC and treating patients accordingly could result in better therapeutic success compared with treating all CRC patients similarly. Studies have highlighted the significance of CRC as a major cause of mortality worldwide and the potential benefits of identifying molecular subtypes to tailor treatment strategies and improve patient outcomes. Methods: This study proposed an unsupervised learning approach using hierarchical clustering and feature selection to identify molecular subtypes and compares its performance with that of conventional methods. The proposed model contained gene expression data from CRC patients obtained from Kaggle and used dimension reduction techniques followed by Z-score-based outlier removal. Agglomerative hierarchy clustering was used to identify molecular subtypes, with a P-value-based approach for feature selection. The performance of the model was evaluated using various classifiers including multilayer perceptron (MLP). Results: The proposed methodology outperformed conventional methods, with the MLP classifier achieving the highest accuracy of 89% after feature selection. The model successfully identified molecular subtypes of CRC and differentiated between different subtypes based on their gene expression profiles. Conclusion: This method could aid in developing tailored therapeutic strategies for CRC patients, although there is a need for further validation and evaluation of its clinical significance.

This study investigates the potential of Explainable Artificial Intelligence (XAI) to enhance mental healthcare by improving diagnostic transparency and treatment precision. Using the “Human Stress Detection” dataset from Kaggle, we applied an ensemble of classifiers, including Random Forest, K-Nearest Neighbors, SVM, XGBoost, and Decision Trees, to assess stress levels based on physiological data. The Random Forest model achieved the highest accuracy of 98.41%. By leveraging XAI techniques like LIME and SHAP, the study demonstrates how XAI can improve the interpretability of mental health assessments, reduce biases, and promote more equitable and effective mental healthcare.

Social media has transformed the way we interact with each other in the twenty-first century. As it continues to evolve and shape our lives, it is important to consider both the benefits and the potential risks of these platforms. One of these risks is the spread of misinformation over global platforms like Twitter. The rapid dissemination of false information can have far-reaching consequences, influencing public opinion and potentially leading to real-world harm. One critical area of concern is the impact of fake news on mental health and well-being. This study aims to explore the emotional responses elicited by fake news and provide insights into the psychological effects of false information circulated through social media. To achieve this, a novel Ensemble Hard Voting Classifier (EHVC), which accurately classifies tweets as either fake or real with an accuracy of 93.84

One common constraint in the practical application of speech recognition is Code Switching. The issue of code-switched languages is especially aggravated in the context of Indian languages – since most massively multilingual models are trained on corpora that are not representative of the diverse set of Indian languages. An associated constraint with such systems is the privacy-intrusive nature of the applications that aim to collate such representative data. To collectively mitigate both problems, this work presents CodeFed: A federated learning-based code-switching detection model that can be deployed to collaboratively be trained by leveraging private data from multiple users, without compromising their privacy. Using a representative low-resource Indic dataset, we demonstrate the superior performance of a collaboratively trained global model that is trained using federated learning on three low-resource Indic languages – Gujarati, Tamil and Telugu and draw a comparison of the model with respect to the most current work in the field. Finally, to evaluate the practical realizability of the proposed system, CodeFed also discusses the system overview of the label generation architecture which may accompany CodeFed’s possible real-time deployment.

The demand for high-quality and diverse 3D content has increased due to its applications in virtual reality, augmented reality, and 3D printing. Converting 2D images to 3D models is a challenging task requiring an understanding of depth, texture, and illumination. This study delves into a novel deep learning-based methodology for generating high-quality and accurate 3D models from 2D images. The proposed approach combines Generative Adversarial Networks (GANs) and Deep Marching Tetrahedra to synthesize complex 3D objects with realistic textures and lighting effects. Additionally, a 2D Texture Generator based on Random Noise and a GAN, as well as a Light Map Generator using Spectral Power Distribution Function and a GAN, are designed to enhance the visual appeal and realism of the generated 3D models. The paper presents a tri-model architecture incorporating a seam-aware landmark detector, which identifies heatmaps to ensure precise mapping of 2D textures onto the 3D model. This feature significantly improves the accuracy and quality of texture application by aligning key points from the 2D images with corresponding areas on the 3D geometry. Furthermore, the model employs point source lighting for light map generation, simulating realistic illumination effects that contribute to the final output’s visual richness. The proposed technique is evaluated, showcasing its superiority over existing methods in generating diverse and realistic 3D models. The study highlights the potential applications of the proposed technique in various domains, including computer graphics, virtual reality, architecture, and industrial design. The ability to generate accurate 3D models with diverse variations opens up exciting opportunities for design exploration and visualization. This work contributes transformative solutions for 3D object synthesis, 2D texture generation, and light map simulation, paving the way for advancements in 3D modeling and design.

Peer-to-peer (P2P) lending is a way to directly connect borrowers and lenders, providing advantages such as quicker access to loans and fewer credit restrictions compared to traditional banking. However, the absence of intermediaries heightens the risk of loan defaults, making robust risk assessment models essential to safeguard the lender's capital. This study suggests a predictive model focused at predicting loan defaults in European P2P lending platform called Bondora, machine learning and deep learning algorithms were used to enhance risk management and mitigate financial losses. The raw data was collected from a European P2P lending platform known as Bondora and it was pre-processed using Z-score normalization to address anomalies and improve accuracy by removing outliers and getting rid of null values. Various machine learning models were implemented such as Decision Tree, Random Forest, and Artificial Neural Networks (ANN). Among the models used the LightGBM model showed the best results by achieving an accuracy of 99.8%. The Decision Tree and Random Forest models achieved comparable accuracy, while the ANN model, which had a higher loss of 1.0848, showed reasonable predictive performance. The findings demonstrated that machine learning models can consistently predict loan defaults, resulting in more informed lending decisions and greater trust among investors in P2P lending. The study found that these models can improve credit risk assessment and provide a feasible alternative to existing risk assessment approaches. Further improvement of these models could result in even more trustworthy predictions, lowering the financial risks involved with peer-to-peer lending.

The quantity of textual data on the internet is growing exponentially, and it is very tough task to obtain important and relevant information from it. An efficient and effective method is required that provides a concise summary of an article. This can be achieved by the usage of automatic text summarization. In this research, the authors suggested an efficient approach for text summarization where an extractive summary is generated from an article. The methodology was modified by integrating a normalized similarity matrix of both BM25+ and conventional TextRank algorithm, which resulted in the improvised results. A graph is generated by taking the sentences in the article as nodes and edge weights as the similarity score between two sentences. The maximum rank nodes are selected, and the summary is extracted. Empirical evaluation of the proposed methodology was analyzed and compared with baseline methods viz. the conventional TextRank algorithm, term frequency–inverse document frequency (TF–IDF) cosine, longest common consequence (LCS), and BM25+ by taking precision, recall, and F1 score as evaluation criteria. ROUGE-1, ROUGE-2, and ROUGE-L scores were calculated for all the methods. The outcomes demonstrate that the proposed method can efficiently summarize any article irrespective of the category it belongs to.

Bitcoin has become a popular method for illegal transactions, such as ransomware payments and money laundering. Detecting these activities within the Bitcoin blockchain is challenging due to the lack of transaction labels and the network's enormous size, allowing bad actors to hide their actions. Previous studies have suggested using unsupervised anomaly detection or supervised and active learning techniques for identifying illicit activity within Bitcoin's network. This paper presents a novel machine-learning methodology that combines feature engineering with supervised learning algorithms to identify illicit transactions in the Bitcoin network. The approach shows promising results in accurately classifying transactions as illicit or legitimate. This method not only provides an efficient solution for detecting unauthorized transactions but also holds significant implications for establishing robust regulatory frameworks for digital currencies.

Soft skills are interpersonal skills that define the personality trait of an individual. These skills influence the social interactions of a person and are described using non-verbal and verbal factors like posture, gesture, vocal tone, etc. Throughout the literature, many attempts have been made to evaluate the soft skills of an individual using computer-based algorithms. Most of these attempts focus on only one of the two factors, i.e., either verbal or non-verbal evaluation. The non-verbal evaluation algorithms mainly consist of lateral body posture evaluation or the study of kinematic behavior using specialized hardware, whereas the verbal evaluation typically includes in-person interviews. Hence, a computerized soft skills evaluation system is required that can automatically evaluate the soft skills of an individual. This study proposes a way to evaluate soft skills using automated frontal posture evaluation and vocal assessment of an individual. The proposed methodology estimates the non-verbal and verbal confidence score of an individual. The non-verbal confidence score is estimated using the combination of the MoveNet-thunder skeleton estimation algorithm and the posture angle evaluation system. This system evaluates the frontal posture of an individual and provides a confidence score based on various posture angles like shoulder alignment, neck bent angle, arm abduction, etc. The verbal confidence score is estimated using pause detection, filler word detection, and continuous word repetition estimation models. The confidence scores generated from these two estimation pipelines are combined to form the overall soft skills confidence score of an individual. This score is compared with a threshold value to validate the results. The threshold value depicts the average and natural soft skills confidence score of an individual. This threshold value was estimated using the natural speech part of the standard Librispeech dataset and the bio-mechanic studies for posture estimation. The proposed model only supports single-user visual input, and it can be improved using cloud implementations and real-time data input considerations.

Cancer is the foremost cause of mortality among humans, as per statistics and accurate classification of the lesion is critical for treating skin cancer at an early stage. Identification of the disease via computer-aided tools can help in accurate diagnosis. This study's primary goal is to suggest an effective strategy for more accurately classifying skin lesions. The binary classification of skin lesions has been proposed using the weighted average ensemble approach. The predictions from various models are combined via the weighted sum ensemble, where the weights of each model are determined by how well it performs. Weights for each learner in the weighted ensemble are scientifically determined based on their average accuracy on the testing dataset. The proposed weighted ensemble classifier uses an ensemble of seven deep-learning neural networks to perform binary classification, including InceptionV3, VGG16, Xception, ResNet50, and others. The International Skin Imaging Collaboration (ISIC) dataset has been used for experimentation, which is binary classified into Melanoma and Nevus. The proposed ensemble method provides the highest level of accuracy, precision, recall, f1-score, sensitivity, and specificity of 93.36%, 93%, 93%, 93%, 97%, and 97% respectively on the first ISIC dataset. The proposed methodology's efficiency has also been compared and evaluated with another ISIC dataset. On the other ISIC dataset, the proposed weighted ensemble classifier had an accuracy of 85.54%. Additionally, the proposed methodology has been compared with state-of-art techniques. a weighted ensemble method where the final result decision is made based on the weighted total of the anticipated outputs from the classifiers. Each model is given a specific weight, which is then multiplied by the value it predicted and used to get the sum or average forecast. The suggested classification model concluded about the expected probabilities for each class and selected the class with the highest probability.

Skin Cancer is one of the most widespread forms of cancer in the world which can be detected using dermatoscopic images. In this paper, a texture based feature extraction algorithm is presented for the classification of dermatoscopic images. A median based Local Ternary Pattern is extracted followed by the computation of local quantized ternary patterns. The feature set extracted is then classified using a modified convolutional neural network. The images used for the detection of multiple types of skin cancer are obtained from two publicly available datasets, HAM10000 and ISICUDA11. For the proposed technique, the average recall value, average precision and average accuracy is found to be 75.20%, 95.44% and 96% respectively. An average increase in accuracy for the proposed algorithm is up-to 50.6%, 24.1% and 4.7% over LTP, DLTerQEP and a DE ANN based algorithm respectively.