A0349
Title: Dense-to-sparse neural network modelling for financial statement data using feature importance attributions
Authors: Lars Fluri - University of Basel (Switzerland) [presenting]
Abstract: A new approach is proposed to feature selection and sparse modelling in the context of financial data analysis to predict free cash flow. Utilising deep learning important features (DeepLIFT), a process for iterative elimination of input features is introduced, reducing the model complexity and enhancing the robustness through the elimination of less significant input nodes. Furthermore, a method for the regrowth of nodes using the gradient magnitude of previously eliminated features is used. Drawing on a dataset of 874 firms from the DACH region over a decade, the model is used to identify forward-looking predictors of free cash flow. Additionally, it evaluates both computational aspects and performance metrics (including in-sample and out-of-sample performance) to measure improvements from the original dense model to the optimized sparse model. The contribution is to the evolving field of machine learning applications in finance, proposing an alternative framework for feature selection and model optimization.
