Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/132748
Title: A systematic study of the performance of Feedforward Neural Networks on data from closed-form mathematical models
Author: Ríos Romero, Harry de los
Tutor: Duch, Jordi  
Abstract: In the age of data, Neural Networks have proven to be powerful predictive tools. However, in scenarios where Model Interpretability prevails over accuracy, other statistical tools may be more appropriate. One such tool is the Bayesian Machine Scientist, a tool based on Bayesian Statistics and Statistical Physics that explores the space of closed-form mathematical models using Markov Chain Monte Carlo and finds the most plausible model given data. In this paper we have made a systematic study of the accuracy of FeedForward Neural Networks, the Bayesian Machine Scientist and a Random Forest algorithm, using the RMSE metric on two types of scenarios. The first one, on data generated by another FeedForward Neural Network, and the second one on a set of data with different noise levels, generated by means of three closed-form mathematical models coming from the Bayesian Machine Scientist. The results show that in the noise-free scenario the Neural Networks are unbeatable but in a real scenario with noise, its performance decreases and could be outperformed by the Bayesian Machine Scientist in the low noise range with the advantage that the latter offers much more interpretable models.
Keywords: bayesian machine
FeedForward
neural networks
Document type: info:eu-repo/semantics/masterThesis
Issue Date: Jun-2021
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es/  
Appears in Collections:Bachelor thesis, research projects, etc.

Files in This Item:
File Description SizeFormat 
hde_los_riosTFM0621memory.pdfMemory of TFM1,69 MBAdobe PDFView/Open