Coreset Construction for Extra Binomial Variation in Binomial Regression

Big data analysis involves additional challenges when applying popular machine learning techniques because of the scalability issue. Hence, some innovative approaches to summarizing huge datasets began to appear such that the summarized data retains much of the information present in the original dataset. These summarized data are called a coreset. Due to the sheer volume of the original dataset, the coreset acts as a trade-off between the space and the information.Prior research works focused on the coreset construction algorithm for logistic regression but not specifically for binomial regression. However, when over-dispersion exists in the original dataset, binomial regression is unsuitable for modeling such datasets. In this article, we propose a coreset construction algorithm that considers the over-dispersion in binomial regression when it exists in the dataset. Apart from experimental results, we also provide associated theoretical results for the proposed coreset algorithm.