Volume 12 | Issue 5 |

 Abstract

This paper presents a novel approach to driver sleep detection using deep learning techniques to enhance road safety. With the increasing number of accidents caused by drowsy driving, an effective detection system is needed. Our framework uses convolutional neural networks (CNNs) to analyze facial expressions, eye movements and head positions captured by an in-vehicle camera While extracting meaningful features from these inputs , the proposed model differentiates driver warning and sleep states accurately in real time. RNNs) are employed to further improve detection performance Extensive tests on various datasets demonstrate the efficiency and robustness of the proposed method under various lighting conditions and driver characteristics enable integration in onboard systems a it already exists without it. Overall, the proposed deep learning-based method provides a practical and reliable solution to enhance road safety by better detecting driver sleep in world conditions in the self-contained.Cambridge Institute of technology Bangalore, India

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Driver Drowsiness Detection Using Deep Learning

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

This paper presents an innovative approach for myocardial infarction (MI) detection through an ensemble of three distinct models: Support Vector Machine (SVM), Random Forest, and Convolutional Neural Network (CNN). Trained on labeled electrocardiogram (ECG) image datasets, each model is individually optimized for effective discrimination between MI and non-MI cases. The models' unique strengths, encompassing SVM's handling of high-dimensional feature spaces, Random Forest's ensemble learning, and CNN's proficiency in hierarchical feature extraction, are strategically combined through the AdaBoost ensemble method. The resulting ensemble model is rigorously evaluated on a separate set of ECG images, demonstrating its enhanced diagnostic accuracy. Key performance metrics, including accuracy, precision, recall, and F1 score, are presented to assess the ensemble model's robustness in real-world clinical applications. This research contributes to the advancement of medical image classification by showcasing the potential of ensemble methods in improving myocardial infarction detection accuracy.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Myocardial infarction detection, Ensemble learning, Support Vector Machine (SVM), Random Forest, Convolutional Neural Network (CNN), AdaBoost, Electrocardiogram (ECG) images, Diagnostic accuracy, Feature extraction

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

In this paper we aim to develop an prediction system using machine learning to detect and classify the presence of diabetes in e-healthcare environment using Ensemble Decision Tree Algorithms for high feature selection. A significant attention has been made to the accurate detection of diabetes which is a big challenge for the research community to develop a diagnosis system to detect diabetes in a successful way in the e-healthcare environment. In this paper we aim to develop an prediction system using machine learning to detect and classify the presence of diabetes in e-healthcare environment using Ensemble Decision Tree Algorithms for high feature selection. A significant attention has been made to the accurate detection of diabetes which is a big challenge for the research community to develop a diagnosis system to detect diabetes in a successful way in the e-healthcare environment. The existing diagnosis systems have some drawbacks, such as high computation time, and low prediction accuracy. To handle these issues, we have proposed diagnosis system using machine learning methods, such as preprocessing of data, feature selection, and classification for the detection of diabetes disease in e- healthcare environment. Model validation and performance evaluation metrics have been used to check the validity of the proposed system. We have proposed a filter method based on the Decision Tree algorithm for highly important feature selection. Two ensemble learning Decision Tree algorithms, such as Ada Boost and Random Forest are also used for feature selection and compared the classifier performance with wrapper based feature selection algorithms also. Machine learning classifier Decision Tree has been used for the classification of healthy and diabetic subjects. The experimental results show that the Decision Tree algorithm based on selected features improves the classification performance of the predictive model and achieved optimal accuracy. Additionally, the proposed system performance is high as compared to the previous state-of-the-art methods. High performance of the proposed method is due to the different combinations of selected features set. Furthermore, the experimental results statistical analysis demonstrated that the proposed method would be effectively detected diabetes disease.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Ensemble learning Decision Tree algorithms, such as Ada Boost and Random Forest

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The assignment "class of site visitors signs and signs and symptoms using deep studying" represents a tremendous expand inside the situation of laptop imaginative and prescient with a completely unique cognizance on the recognition and kind of site visitors signs and symptoms. We used the energy of Python to resolve a complex visitors signal class trouble the usage of prominent models: the MobileNet and YOLOv5 architectures. The MobileNet structure done vast tiers of typical overall performance with a schooling accuracy of 97.00% and a validation accuracy of 98.00%. The quit end result is a set of four,100 seventy carefully curated snap shots overlaying fifty eight education of numerous road symptoms, which incorporates speed limits, site visitors signs and symptoms, prohibition symptoms and signs, threat warnings, and additional. the ones hours cover the overall range of site visitors rules and offer complete coverage of what is going to be stated. The YOLOv5 implementation introduced real-time road sign reputation using actual-time pictures and webcam statistics. The version changed into professional on a dataset containing 39 specific website online visitors sign instructions. those instructions encompass a enormous style of signs and signs and symptoms at the side of pedestrians, pace limits, warning and regulatory signs, and help you observe your mission to actual-international

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Traffic Sign Recognition, Neural Network Architecture, Object Detection

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

This paper proposes a novel approach utilizing machine-gaining knowledge of strategies and Xception architecture for PD detection, focusing on spiral and wave drawings, common diagnostic tools in clinical practice. Through a dataset collection process, including individuals with and without PD, preprocessed data were employed to train machine learning models. Results indicate promising performance, demonstrating the potential of machine learning and Xception architecture in early PD detection. This approach offers advanced accuracy and efficiency in diagnosis, ultimately leading to better patient outcomes and enhanced quality of life.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Parkinson's disease, Neurodegenerative disorder, Machine learning, Xception architecture, Early detection, Diagnosis, Spiral and wave drawings, Clinical practice, Image classification, Convolutional neural networks, Depthwise separable convolutions, Inception modules, Model performance

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

This study provides an in-depth examination of sophisticated machine learning techniques for predicting brain strokes using the Healthcare Dataset Stroke Data. Brain stroke prediction is a critical task in healthcare, having the capacity to greatly enhance patient outcomes via early identification and intervention. In this study, We evaluate the effectiveness of four cutting-edge algorithms: Convolution-Based Neural network(CNN), CNN with Long Short-Term Memory (CNN-LSTM) architecture, XGBoost, and LightGBM. We evaluate these algorithms based on their predictive accuracy, sensitivity, specificity, and computational efficiency. Our research clarifies the advantages and disadvantages of each algorithm in the context of brain stroke prediction, providing valuable insights for healthcare practitioners and researchers seeking to leverage machine learning for early stroke detection. This dataset is used to predict whether a patient is likely to get stroke based on the input parameters like gender, age, and various diseases and smoking status. A subset of the original train data is taken using the filtering method for ML and Data Visualization purposes.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Brain stroke prediction, XGBoost, LightGBM, Convolution neural networks (CNN), CNN-LSTM, Early stroke detection, Data visualization, healthcare stroke dataset.

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The increasing popularity of the electric vehicles(EVs) has spurred the need for best battery thermal management systems to ensure optimal performance, longevity, and safety of energy storage systems. It focuses on leveraging Artificial Intelligence (AI) technologies, specifically the Multilayer Perceptron (MLP) algorithm, to enhance the efficiency of battery thermal management in EVs. I algorithms, such as MLP,offers the potential to model and predict the thermal behavior of batteries more accurately, allowing for real- time adjustments and improved control strategies. With the large-scale commercialization and growing market share of electric vehicles (EVs). Their focus has been on higher energy efficiency, an improved thermal performance, and optimized multi- material battery enclosure designs. The combination of simulation-based design optimize the battery pack and Battery Management-System (BMS) is evolving and has expanded to include novelties such as artificialintelligence/machine learning (AI/ML) to improveefficiencies in design, manufacturing, and operations for their application in EVs and energy storage systems. Specific to BMS, these advanced concepts enable a more accurate prediction of battery performance such as its State of Health(SOH), State of Charge(SOC), and State of Power(SOP). This study presents a comprehensive evaluation of the latest developments and technologies in battery design, thermal management, and the applicationof AI in Battery Management Systems(BMS) for electric vehicles (EVs).

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Battery Thermal Management in EV Using AI

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

With the growing use of authentication systems in the recent years, fingerprint spoof detection has become increasingly important. In this model, we use Convolutional Neural Networks (CNN) for fingerprint spoof detection. Our system is trained on the datasets used in The Liveness Detection Competition of years 2009, 2011 and 2013, which comprise almost 50,000 real and fake fingerprints images. The CNN is pre-trained on natural images and fine-tuned with the fingerprint images, CCN with random weights, and a classical Local Binary Pattern approach. The project shows that pretrained CNNs can yield state-of-the-art results with no need for architecture or hyperparameter selection. Dataset Augmentation is used to increase the classifiers performance, not only for deep architectures but also for shallow ones. We also report good accuracy on very small training sets (400 samples) using these large pre-trained networks. The model achieves an overall rate of 97.1% of correctly classified samples - a relative improvement of 16% in test error when compared with the best previously published results

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Fingerprint recognition, Feature extraction, Convolutional neural network

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

The "Converting podcast episodes into text format and summarizing them" project aims to automate the process of summarizing podcasts, making it convenient for users to quickly grasp the key points of lengthy audio content. The process begins with data collection, where podcast episodes are gathered either as transcripts or audio files. For transcripts, preprocessing techniques are applied to clean the text data, removing unnecessary characters and tags. For audio files, speech recognition tools are employed to convert spoken words into text. The summarization techniques primarily include both extractive and abstractive methods.

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

CONVERTING PODCAST EPISODES INTO TEXT FORMAT AND SUMMARIZING THEM

  License

Creative Commons Attribution 4.0 and The Open Definition

 Abstract

Automatically identifying animal species in images is vital for ecology, conservation, and biodiversity studies. Deep learning, particularly convolutional neural networks (CNNs), has become a powerful tool for this task. We compared five pre-trained CNN models (AlexNet, VGG16, VGG19, ResNet50, InceptionV3) on a dataset of 20 animal species with 19 classes from KTH-Animal dataset and one class from Kaggle dataset. Our approach involved fine-tuning these models with pre-extracted features. We evaluated accuracy, precision, recall, F1 score, false acceptance rate (FAR), and false rejection rate (FRR).VGG16 achieved the highest accuracy (95.73%) and F1 score (0.94), excelling at correctly identifying animals with minimal misclassifications (FAR and FRR of 5% each). InceptionV3 followed closely (94.51% accuracy, 0.95 F1 score). AlexNet and ResNet50 showed a trade-off between precision and recall, making them potentially useful for specific needs. This study highlights the effectiveness of pre-trained features in CNNs for animal species recognition, especially after fine-tuning. This approach reduces reliance on large, labeled datasets, making it valuable for ecological applications with limited data. Our VGG16-based approach outperforms previous works, showcasing the potential of deep learning for animal species recognition

Licence: creative commons attribution 4.0

  License

Creative Commons Attribution 4.0 and The Open Definition

 Keywords

Animal species recognition, deep convolutional neural networks, transfer learning, camera-trap, KTH dataset.

  License

Creative Commons Attribution 4.0 and The Open Definition