Model to Predict Incoming Tech Support Demand in a Banking Company Applying CRISP-DM and Machine Learning

Abstract

In modern-day businesses, forecast systems had become into a valuable tool in areas such as finances, advertisement, healthcare and more. The amount of data available online, make it possible to analyze and prevent future customer and market behaviors. The goal of the study is to train machine learning models to predict incoming tech support demand. The algorithms selected in this study are: Random Forest, Multilayer Perceptron and Long Short-Term Memory. The methodology used in this study is Cross Industry Standard Process for Data Mining, which comprises the following phases: business understanding, data understanding, data preparation, modeling, and evaluation. The dataset gathered was comprised by 17,847 tech support issue tickets, collected between January 2020 and May 2023 (Ten temporality variables were identified, with the variable 'year' standing out as the most relevant within this specific dataset). The amount of data endowed the models with adaptability and accuracy when generating predictions. The results obtained showed that the Random Forest algorithm achieved an R2 metric of 0.80, positioning it as the technique that exhibited the best fit with the study's dataset.