Predictive models trained on aggregated data

The aim of this work is to expand the range of methods trained on aggregated data, which are combinations of all possible attribute pairs. The work compares the results obtained by logistic regression models and the naive Bayes classifier for different datasets based on fine-grained data with the results achieved by modified models based on the aforementioned aggregates. Furthermore, this work presents and examines a new method called RanAgg, based on aggregated data. As part of the research described in this work, the properties of the mentioned methods and the impact of their individual parameters on the speed and quality of predictions based on defined measures were investigated. The results obtained indicate that the proposed modified logistic regression models and the naive Bayes classifier, as well as the new RanAgg method, allow for prediction based solely on aggregated data.