Fourier-Enhanced Neural Networks: An Approach to Predicting Business Key Performance Indicators

This paper introduces Fourier-Enhanced Neural Networks (FENN) to improve Marketing Mix Modelling (MMM) for predicting key performance indicators (KPIs) in a Central and Eastern Europe (CEE) region company. Traditional MMM methods often struggle with complex temporal patterns in business data. The FENN methodology integrates Fourier Transform techniques with neural network architectures, using a one-to-one connection matrix as a weight multiplier to decompose time series data into frequency components, enhancing the model's ability to detect periodicities and trends. The FENN methodology is detailed, including neural network layer multiplications and ReLU activation functions to model seasonality and media investments. The dataset is divided into training, testing, and validation splits, optimizing the Fourier order based on the validation set. Results include visualizations of time series decomposition and seasonality estimates using Fourier methods versus standard seasonal variables. The findings show that FENN improves KPI prediction accuracy, demonstrating the robustness and utility of Fourier-enhanced neural Networks for optimizing marketing strategies and supporting data-driven decision-making.