The COVID-19 pandemic presented unprecedented challenges to global economies, revealing vulnerabilities and emphasizing the urgent need for predictive tools to support timely and informed decision-making. In this context, this study investigates the application of Multilayer Perceptron (MLP) artificial neural networks (ANN) to forecast four key economic indicators: Gross Domestic Product (GDP), unemployment rate, inflation rate, and tourism, in six European countries (Austria, France, Germany, Italy, Portugal, and Spain). Based on a time series of public health and economic data collected between January 2020 and May 2023. The methodology involved developing two forecasting models: (1) a joint model trained to predict all indicators simultaneously, and (2) independent models, where each economic indicator was forecasted using a dedicated network. Data preparation included population-adjusted normalization and country-specific normalization for GDP, followed by a sliding window approach using 30-day input periods. The dataset was split into training, validation, and testing subsets, and experiments were conducted with varying activation functions and hidden layer configurations within the MLP networks. The results show that both approaches achieved high predictive accuracy, with independent models performing better in forecasting GDP and Inflation, while the joint model delivered superior results for unemployment and tourism. The lowest mean absolute error (MAE) was recorded for tourism (0.0013) using an independent network, and the joint model achieved an MAE of 0.0018 for unemployment. These findings highlight that network architecture should be tailored to specific forecasting objectives, as different economic indicators benefit from distinct configurations. The study reinforces the potential of MLP networks as effective tools for economic forecasting during crisis periods.
Monteiro et al. (Thu,) studied this question.