We present a neural network model for estimation of multiple conditional quantiles that satisfies the noncrossing property. Motivated by linear noncrossing quantile regression, we propose a noncrossing quantile neural network model with inequality constraints. In particular, to use the first-order optimization method, we develop a new algorithm for fitting the proposed model. This algorithm gives a nearly optimal solution without the projected gradient step that requires polynomial computation time. We compare the performance of our proposed model with that of existing neural network models on simulated and real precipitation data. Supplementary materials for this article are available online.

Statistical models that can generate a road-traffic noise map for a city or area where only elementary urban design factors are determined, and where no concrete urban morphology, including buildings and roads, is given, can provide basic but essential information for developing a quiet and sustainable city. Long-term cost-effective measures for a quiet urban area can be considered at early city planning stages by using the statistical road-traffic noise map. An artificial neural network (ANN) and an ordinary least squares (OLS) model were developed by utilizing data on urban form indicators, based on a 3D urban model and road-traffic noise levels from a normal noise map of city A (Gwangju). The developed ANN and OLS models were applied to city B (Cheongju), and the resultant statistical noise map of city B was compared to an existing normal road-traffic noise map of city B. The urban form indicators that showed multi-collinearity were excluded by the OLS model, and among the remaining urban forms, road-related urban form indicators such as traffic volume and road area density were found to be important variables to predict the road-traffic noise level and to design a quiet city. Comparisons of the statistical ANN and OLS noise maps with the normal noise map showed that the OLS model tends to under-estimate road-traffic noise levels, and the ANN model tends to over-estimate them.

This study aimed to screen the inhalation toxicity of chemicals found in consumer products such as air fresheners, fragrances, and anti-fogging agents submitted to K-REACH using machine learning models. We manually curated inhalation toxicity data based on OECD test guideline 403 (Acute inhalation), 412 (Sub-acute inhalation), and 413 (Sub-chronic inhalation) for 1709 chemicals from the OECD eChemPortal database. Machine learning models were trained using ten algorithms, along with four molecular fingerprints (MACCS, Morgan, Topo, RDKit) and molecular descriptors, achieving F1 scores ranging from 51 % to 91 % in test dataset. Leveraging the high-performing models, we conducted a virtual screening of chemicals, initially applying them to data-rich chemicals generally used in occupational settings to determine the prediction uncertainty. Results showed high sensitivity (75 %) but low specificity (23 %), suggesting that our models can contribute to conservative screening of chemicals. Subsequently, we applied the models to consumer product chemicals, identifying 79 as of high concern. Most of the prioritized chemicals lacked GHS classifications related to inhalation toxicity, even though they were predicted to be used in many consumer products. This study highlights a potential regulatory blind spot concerning the inhalation risk of consumer product chemicals while also indicating the potential of artificial intelligence (AI) models to aid in prioritizing chemicals at the screening level.

Recent advances in a generative neural network model extend the development of data augmentation methods. However, the augmentation methods based on the modern generative models fail to achieve notable improvement in class imbalance data compared to the conventional model, Synthetic Minority Oversampling Technique (SMOTE). We investigate the problem of the generative model for imbalanced classification and introduce a framework to enhance the SMOTE algorithm using Variational Autoencoders (VAE s). Our approach systematically quantifies the density of data points in a low-dimensional latent space using the VAE, simultaneously incorporating information on class labels and classification difficulty. Then, the data points potentially degrading the augmentation are systematically excluded, and the neighboring observations are directly augmented on the data space. Empirical studies on several imbalanced datasets represent that this simple process innovatively improves the conventional SMOTE algorithm over the deep learning models. Consequently, we conclude that the selection of minority data and the interpolation in the data space are beneficial for imbalanced classification problems with a relatively small number of data points.