Unveiling Recurring Financial Patterns: Novel unsupervised filtering algorithms for enhanced forecasting

Álvaro Ibrain, Verónica Hernández, Luis Peinado

Abstract: In the domain of predicting forthcoming financial transactions for retail customers, the challenge extends beyond forecasting the monetary amount alone; it requires the anticipation of the next date of occurrence, given the absence of a fixed known periodicity. This research focuses on the identification of recurring transactional patterns, addressing the challenge through the introduction of an innovative set of preprocessing algorithms. The proposed techniques simplify the subsequent "next date" and "next amount" forecasting tasks by disentangling overlapping patterns and eliminating non-recurrent data points. Given the unsupervised nature of this problem, attributed to the absence of labeled data indicating whether a movement is part of a recurrent pattern or not, a series of experiments is conducted to demonstrate the efficacy and robustness of the proposed methodology.