Logistic Regression Models for Aggregated Data

Logistic regression models are a popular and effective method to predict the probability of categorical response data. However, inference for these can become computationally prohibitive large datasets. Here we adapt ideas from symbolic data analysis summarize collection predictor variables into histogram form, perform on this summary dataset. We develop based composite likelihoods derive an efficient one-versus-rest approximate likelihood model histogram-based random variables, constructed low-dimensional marginal histograms obtained full histogram. demonstrate that procedure achieve comparable classification rates standard multinomial against state-of-the-art subsampling algorithms logistic regression, but at substantially lower computational cost. Performance is explored through simulated examples, analyses supersymmetry satellite crop Supplementary materials article available online.