Abstract
The financial sector, and specifically banking entities, have experienced changes in recent years thanks to technology, such as the digitization of transactions and the creation of applications such as digital wallets and PFM (Personal Finance Manager), generating gigabytes of information. Managing knowledge becomes essential to face new competitors, provide better services, understand the financial behavior of clients and face great challenges when processing and analyzing the volume of information available, which in most cases requires a complex preprocessing process and data quality. This is the case of banking transactions, which include free text information in their observation fields, making analysis and classification difficult, preventing the bank and its clients from analyzing financial behavior over a period of time. To solve this problem, the use of Machine Learning techniques was proposed to automate the transaction classification process based on text written in natural language, and provide the information that allows an analysis of the financial behavior and personal expenses of each user. Once the training, evaluation and comparison of different models was completed, using the CRISP-DM methodology as a development framework, an optimized solution was reached that solves the classification problem using the KNN algorithm, with an accuracy close to 96%. The results showed a high level of confidence when classifying a transaction, based on a description, into a category.
Publisher
Salud, Ciencia y Tecnologia