Comparison of stamp classification using SVM and random ferns

In distributed software systems and processes that use large amounts of documents there is an essential need for data mining and document classification algorithms. These algorithms are aimed at optimizing the process, making it less error prone. In this paper we deal with the problem of document classification using two machine learning algorithms. Both algorithms use stamp images in documents to classify the document itself. The idea is to classify the document stamp and then, using known information about the stamp owner, search the rest of the document for relevant data. Our results are based on actual documents used in the process of debt collection and our training and test datasets are randomly picked from an existing database with over three million documents. The mentioned machine learning classification algorithms are implemented and compared in terms of classification accurateness, robustness and speed.