Transaction categorisation is a process of identifying the context or purpose of a bank account transaction, based on its description, transaction amount, date, and contextual metadata. Combining machine learning algorithms with experts’ know-how, Nordigen has built an engine that outperforms both manual bank-statement analysis done by credit specialists (by eliminating human error and optimizing time consumption) and machine learning-only engines, traditionally used by Personal Finance Management (PFM) applications, which have higher error rate due to poor training data quality.
The granularity of the category database caters a broad range of applications. Precise transaction categorisation provides data teams with high quality data for analytics. Generally Nordigen uses a three tiered hierarchical taxonomy, in which each transaction is described at least with one and at most with three level granularity, always assigning lowest possible level. Taxonomy is described in depth in category tree section.
From the perspective of data science, Nordigen's transaction categorisation is a hybrid model that tackles single-label multi-class classification task - the engine assigns one category to a string of characters (i.e. transaction).