Pulmonary diseases, such as chronic obstructive pulmonary disease(COPD) and asthma, are among the leading causes of death in the US.These lung diseases often are diagnosed by pulmonologists using physical exam (e.g., lung auscultation) and objective measurement of lung function with pulmonary function testing. These extensive tests, often happen at a later stage when patients have shown more progression and can be inaccessible to many patients due to limited resources and availability. Nowadays hand-held medical devices (e.g., electrocar-diogram (ECG) monitors) are already available to such patients and can yield ECG data that potentially could be used for diagnosis. To this end, we introduce EPDD: an ECG-based pulmonary disease detector using machine learning to detect pulmonary disease. In this paper, we focus on one particular lung disease called obstructive lung disease (OLD) to explore the use of easily accessible ECGs to train machine learning models to classify whether a patient has normal or severe OLD. Not only do we utilize the time-series raw ECG directly, we also define and extract features from the PQRST waves of the ECGs. We propose to use traditional machine learning models (such as Random Forest, Logistic Regression, XG Boost, Support Vector Machine, K-Nearest Neighbors, and ensemble of them), as well as deep learning models (Convolutional Neural Network, Long Short-Term Memory, and Vision Transformer). Using a dataset of ECGs we collected from 11,346 patients, our experiments show that all of our proposed models significantly outperform our two baseline models - Visual Geometry Group-16 (VGG-16) and Residual Network-50 (ResNet-50) trained on ECG images generated by various signal processing methods, and that the traditional models trained on extracted features from ECGs outperform the deep learning models trained on ECGs and extracted features. Finally, with the goal of providing accessible and affordable healthcare, we design and develop a mobile app based prototype of EPDD to visualize predictions of the selected trained models to patients and pulmonologists.

Knowing the Remaining Useful Life (RUL) of aircraft engines is of paramount importance in the aviation industry. RUL helps anticipate engine failures beforehand so that airlines can proactively schedule maintenance, optimize resource allocation, and reduce the risk of downtime. In this work, we consider two problems on the prediction of RUL: a binary classification problem to predict whether the engine will fail within a month, and a regression problem to predict the remaining number of operational cycles before engine failure. To this end, using the NASA C-MAPSS dataset, we trained a slew of machine learning models to address the aforementioned two problems. Our results show that the Long Short-Term Memory (LSTM) model performed the best on the binary classification problem with 0.95 precision, 0.88 recall and 0.91 F1-score on the test set, and that the Convolutional Neural Network (CNN) model [1] best on the regression problem with 14.02 RMSE on the test set. The state-of-the-art paper documented an RMSE of 16.42. Our work not only surpassed the reference RMSE but also demonstrated superior predictive accuracy. Comparing these results highlights the substantial progress achieved in predicting aircraft engine Remaining Useful Life, showcasing the effectiveness of the models developed in this study in outperforming the state-of-the-art benchmark.

Atmospheric particulate matter (PM), such as PMIO and PM2.5, significantly impacts ecosystems, climate, and human health. Previous studies have shown that black carbon, a type of PM2.5, significantly contributes to global warming. To this end, this study introduces a set of advanced neural network models aimed at improving air quality predictions within ur-ban environments, specifically targeting PM2.5 and PMIO. By harnessing data from Chile's National Air Quality Information System (SINCA), we employ a diverse array of neural networks, including convolutional neural networks (CNNs), long short-term memory networks (LSTMs), a hybrid CNN-LSTM model called ConvLSTM, transformers, and XGBoost models. Our empirical results show that these models are adept at forecasting air pollution trends, with CNN in particular outperforming all other models, providing crucial insights for real-time environmental monitoring. Our innovative approach enables policymakers and public health officials to take timely and effective measures to combat urban air pollution, thereby enhancing the overall quality of life in metropolitan areas.

Media haze (MH) is a serious eye condition that can lead to various issues. This is especially true since MH is often an early warning sign of more severe conditions. If not treated quickly, it can potentially cause discomfort, blurred vision, or even blindness. Currently, MH is often detected by manually examining fundus images. This process allows for human error, where MH may not be detected. This study attempts to increase the detection and diagnosis of MH through our methods of preprocessing fundus images being examined, as well as the use of machine learning models. The models examined include LeNet-5, MLeNet (a modified LeNet), AlexNet, MobileNet, ResNet152v2, and VGG-16, where the former three models were trained on the RFMiD dataset, and the latter three were trained using pre-trained models fine-tuned on RFMiD. We found that all models using the proposed preprocessing demonstrated better performance than using the original images. The most effective model in this study was MobileNet, achieving 97.2% accuracy and 99.2% AUC score on the test set, outperforming the state-of-the-art model we found in the literature.

Breast density is an indicator of a patient’s predisposed risk of breast cancer. Although not fully understood, increased breast density increases the likelihood of developing breast cancer. Accurate assessment of breast density from mammogram images is a challenging task for the radiologist. A patient’s breast density is assigned to one of four categories outlined by Breast Imaging and Reporting Data Systems (BIRADS). There have been efforts to identify automated approaches to assist radiologists in the classification of a patient’s breast density. The interest in using deep learning to fulfill this need for an automated approach has seen a significant increase in recent years. The preprocessing techniques used to develop these deep learning approaches often have a profound impact on the model’s accuracy and clinical viability. In this paper, we outline a novel image preprocessing technique where we concatenate individual mammogram images and compare the results using this technique between Inception-v3 and a vision transformer (ViT). The results are compared using the area under (AUC) the receiver operator characteristics (ROC) curves and traditional accuracy metrics.

Recent years have witnessed continuous growth of the abilty of convolutional neural networks (CNNs) on solving image classification problems. More recently, CNNs have started to show promising results on classifying recyclable materials given their images. This work aims to showcase the effectiveness of CNN architectures for such classification task on our dataset combined from multiple public sources, where in total 12,873 images of recyclable materials are collected over four classes: glass, metal, paper, and plastic. To the best of our knowledge, our work is the first attempt to train and test CNN models on this dataset. To this end, we train a selection of CNN models, including a simple 8-layer CNN, AlexNet, VGGNet, and InceptionNet. Our empirical results show that VGG-16 (a VGGNet variant of 16 convolutional and pooling layers), combined with transfer learning, produces the best testing accuracy of 84.6%. Furthermore, we import this best model to implement a Raspberry Pi application and an Android application to demonstrate their potential for consumer and industrial usage.

In this work, we consider distance-based clustering of partial lexicographic preference trees (PLP-trees), intuitive and compact graphical representations of user preferences over multi-valued attributes. To compute distances between PLP-trees, we propose a polynomial time algorithm that computes Kendall's Tau distance directly from the trees and show its efficacy compared to the brute-force algorithm. To this end, we implement several clustering methods (i.e., spectral clustering, affinity propagation, and agglomerative nesting) augmented by our distance algorithm, experiment with clustering of up to 10,000 PLP-trees, and show the effectiveness of the clustering methods and visualizations of their results.

We introduce partial lexicographic preference trees (PLPtrees) as a formalism for compact representations of preferences over combinatorial domains. Our main results concern the problem of passive learning of PLP-trees. Specifically, forseveral classes of PLP-trees, we study how to learn (i) a PLP-tree consistent with a dataset of examples, possibly subject to requirements on the size of the tree, and (ii) a PLP-tree correctly ordering as many of the examples as possible in case the dataset of examples is inconsistent. We establish complexity of these problems and, in all cases where the problem is in the class P, propose polynomial time algorithms.