This Shiny app illustrates the optimal homogeneous prediction methodology developed by Benjamin Bobbia and Stilian Stoev. It allows the user to upload training and test datasets, select a response variable and predictor variables, and compute an estimate of the optimal homogeneous predictor via quantile random forests using the R-package 'ranger'. The predictions are graphed over the test data set and various metrics are displayed in a table such as TSS (True Skill Statistic), Precision, F1 Score, and Miss Rate. The empirical tail-dependence of the response and the predictor are illustrated in another tab. NOTE: By default the App loads a training and testing data frames from daily maxima of X-ray flux, and 23 other predictors of the so-called SHARP time series derived from active regions of the Sun compiled and curated by Victor Verma. The prediction of Solar flares is challenging, so the please do not be surprised with the low scores. Try it with your own data and let us know how it goes! (c) {sstoev,vkverma}@umich.edu and Benjamin.BOBBIA@isae-supaero.fr.
Color coding:
True Y vs Predicted Y scatter plot. Alarm and event thresholds shown.
Download the predicted values alongside the true values for the selected response variable.
Download CSV