Comparison of Models for Predicting the Number of Calls Received in a Call Center through Time Series Analysis

Abstract

Time series analysis is a crucial aspect of machine learning that deals with data points ordered by time. Time series data is prevalent in various domains, including finance, economics, healthcare, weather forecasting, and many others. Understanding and modeling time series data is essential for making predictions, identifying trends, and extracting meaningful insights. Effectively modeling time series data is a complex task that requires a combination of statistical methods, machine learning algorithms, and domain-specific knowledge. The choice of a specific model depends on the characteristics of the data and the goals of the analysis or prediction task. Our research provides an innovative method to carry out the analysis of time series data. This method is based on successive sequential steps to perform the temporal analysis. Each step is explained theoretically, and then tested on real data. Furthermore, we apply and compare different models based on both statistical approaches, i.e., Seasonal Autoregressive Integrated Moving Average (SARIMA), Seasonal Autoregressive Integrated Moving Average + Exogenous Variables (SARIMAX), and neural networks, i.e., Long Short-Term Memory (LSTM). For the comparison between the models, the Mean Absolute Error (MAE) and Root Mean Squared Error (RMSE) metrics are used, as well as another empirical metric provided by the collaborating company. In each methodology, two verticals are defined, one with exogenous variables and the other without them. The conclusions of the study show that considering the nature of the data analyzed, the model based on neural networks using exogenous variables is the one that provides the best results.