This study investigates the application of machine learning to enhance the accuracy of saturated flow boiling heat transfer correlations. Traditional correlations often exhibit limitations, motivating the exploration of alternative approaches. A comprehensive dataset encompassing 2770 experimental observations, compiled from over 15 published research papers, serves as the foundation for this research. The data incorporates flow boiling heat transfer for pure and mixture refrigerants in smooth tubes. The study compares three machine learning models including neural network, linear regression, and SVM. The data is strategically divided: 80% for training the models and 20% for testing. The wide neural network shows that the best-performing model achieves a Root Mean Square Error (RMSE) of 0.202, demonstrating exceptional prediction accuracy, followed by the linear regression, and SVM training models. Heat Transfer Coefficient (HTC) is predicted with an error of less than 5% for all refrigerants, further confirmed by a high coefficient of determination (0.99). This signifies a significant improvement in traditional experimental-based correlations. Furthermore, the study demonstrates the efficacy of machine learning models in predicting heat transfer enhancement methods. In conclusion, this work highlights the potential of machine learning for refining saturated flow boiling heat transfer predictions.

This study delves into the realm of healthcare data analysis within the framework of local differential privacy (LDP), aiming to assess the performance of various methodologies under this stringent privacy paradigm. Leveraging LDP, which ensures individual data privacy while allowing for analysis, we explore the landscape of healthcare data with a focus on meaningful performance evaluation. Through a systematic investigation, this research evaluates the performance and efficiency of different approaches, shedding light on their suitability for healthcare data applications. By synthesizing insights from diverse techniques and considering their implications within the context of LDP, this study contributes to the advancement of privacy-preserving healthcare analytics while maintaining the integrity and utility of the underlying data. Our findings indicate that Reduced and Castell approaches present the best performance while maintaining a low epsilon, ε < 1. The Castell approach, with its lower memory consumption than the Reduced approach, stands out as the best approach for large healthcare datasets.

Nowadays, our personal information is highly values with companies, hospitals, and internet services, among other industries, using it to create databases to extract user statistics or to improve their services. Local Differential Privacy (LDP) studies how these services can use the information while preserving users’ privacy through various techniques. LDP approaches have been proposed to preserve the privacy of databases and provide statistical approximation, but they are limited when dealing with k-dimensional distribution estimations. To address this drawback, we propose applying Bayesian ridge regression to the central server to recover the original distribution. The propose method has been tested on three open datasets with varying characteristics, including the number of users and the list of attributes, as well as their cardinality. Further, the proposed method outperformed the well-known LoPub and LoCop algorithms. With a privacy budget set at 0.16 per attribute and k-way set at 5, our work achieves a 57% decrease in the average variant distance compared to that of LoPub and a 50% decrease compared to LoCop. Our results suggest that a Bayesian ridge algorithm can be a useful tool for privacy preservation during data publication and it may have various applications where privacy is a concern.

This work studies a vulnerability in face obfuscation techniques intended to preserve the privacy of individuals. There have been several attempts to prevent unauthorized face recognition from being performed, aiming to guarantee anonymity in video data. Most of these attempts have focused on facial areas that are thought as sensitive to contribute most to facial recognition. However, obfuscation of such facial areas is insufficient to preserve privacy because gait information such as arm movements and step characteristics can be used to identify individuals and other personal information such as gender. In this paper, we claim that individual tracking and gender estimation are possible just from the gait information extracted from a video without using face-related data. We propose a set of biometric features and an algorithm to estimate gender from skeleton data. Our experiments with more than 100 subjects demonstrate that gender is estimated with a significant accuracy of 99.86%. The proposed identification algorithm, which is based on pattern-matching techniques, is robust against changes in the manner of walking and successfully identifies subjects with only small error of 0.036.

Heart disease is the principal cause of mortality and the major contributor to reduced quality of life. The electrocardiogram is used to monitor the cardiovascular system. The correct classification of the beats in electrocardiograms gives an opportunity to have treatment more focused. The manual analysis of the ECG signals faces different problems. For this reason, automated diagnosis systems are fed by ECG signals to detect anomalies. In this paper, we propose a method based on a novel preprocessing approach and neural networks for the classification of heartbeats which is able to classify five categories of arrhythmias in accordance with the AAMI standard. The preprocessing stage allows each beat to have “P wave-R peak-R peak” information. We evaluated the proposed method on the MIT-BIH database, which is one of the most used databases. According to the results, the proposed approach is able to make predictions with the average accuracies of 97%. The average accuracies are compared to different approaches that use different preprocessing and classifier stages. Our approach is superior to that of most of them.

The facial expression recognition has been a topic of active researches given a result the proposal of several efficient algorithms; however, in most cases they remain limited to controlled conditions situations. In this study, we tackle the challenge of recognizing emotions through the facial expression into activities in-the-wild adding the accuracy rate for each expression. To this end we an algorithm that allows accurate face expression recognition in an uncontrolled environment, that means different kind of illumination, backgrounds, occlusions, face's profiles, etc. Proposed system firstly detects different profile of face (left, frontal and right), Then it uses only the frames in which the face profile is frontal, in the next step the face regions of interest (ROI) are segmented automatically to carry out the feature extraction. We use a classifier based on clustering, it has the advantage that if a new class (emotion) is added, it is not necessary to train this completely. Proposed system was evaluated using short video clips of several pictures together with description sentences describing the main activity in the video. The evaluation results show that the proposed scheme is able to recognize the face's profiles with the recognition rate to approximately 93% and principal emotions in unconstrained video sequences.