Ozone air concentration trend attributes assist hours-ahead forecasts from univariate recorded data avoiding exogenous data inputs

Machine learning for hours-ahead forecasts of urban air concentrations of oxides of nitrogen from univariate data exploiting trend attributes

“…A recently developed univariate trend-attribute method has successfully been developed and applied to multi-year hourly ozone concentrations to predict hours-ahead ozone concentrations in city air. 27 This approach is adapted in this study to predict hours-ahead NOx concentrations. NOx hourly trends in city air tend to be more spiky than those recorded for ozone because anthropogenic influences have more immediate impacts.…”

“…This study adapts the recently proposed trend-attribute time-series analysis applied to predict hourly ozone air levels to generate ML models for short-term NOx predictions at urban recording sites in eight cities in Central England from 2017 to 2021. 27 It compares the distinct NOx prediction requirements of urban background and roadside recording sites and the impact of reduced NOx concentrations in 2020 and 2021 related to COVID-19 lockdowns. The relative importance of specific trend attributes calculated with data from the prior twelve hours to NOx forecasts for specific hours ahead is also established.…”

Spatio-Temporal Attributes of Varicella-Zoster Case Number Trends Assist with Optimizing Machine Learning Predictions

“…It does this by applying multiple repeat runs with distinct data record combinations to provide statistical confidence in terms of mean and standard deviations of error metrics. The recently developed multi-K-fold cross-validation technique [26] extends basic crossvalidation to repeatedly perform such "leave-one-set-out" analysis with a sequence of different percentage splits (e.g., 3fold in each run rotates three times through 1/3 of the data records randomly assigned to validation and 2/3 assigned to training; 15-fold rotates 15 times through 1/15 of the data records assigned to validation and 14/15 assigned to training). In this study, 3-fold, 4-fold, 5-fold, 10-fold, and 15-fold analysis was conducted and repeated in multiple leave-one-set-out trials.…”