Financial forecasting has emerged as a key field of study in the last decades as budget
planning, and decision-making processes play a vital part in establishing and
maintaining a healthy and sturdy business. Last years, Machine Learning and Deep
Learning’s concepts have become prominent in the financial industry due to their
ability to handle vast amounts of data and extract time-dependent patterns from
non-linear relationships along with the -nowadays- exponentially growing availability
of computing power. More and more industries and business operations tend to
deploy artificial intelligence technology to automize their processes, minimize their
risks and optimize their development in terms of increasing their revenues as a
consequence of high quality and robust productivity. The wide range of Deep
Learning usage in the finance industry extends from security and fraud detection,
underwriting, stock marketing predictions, and chatbot advisory. [1][2][3]
The possible outcomes and variations of Machine Learning applications are fully
capable to cover a wide spectrum of needs and ideas and in this work we have
developed a theoretical framework to acquire fundamental understanding of the Deep
Learning philosophy and concepts, studying a very popular Neural Network type by
the name of Recurrent Neural Networks, and another promising one by the name
Reservoir Computing and their predictive abilities on time series of data. In addition
to this, our work aims to study systematically and evaluate different ways to convert a
dataset into a more «friendly» form, to optimize the predicting ability of the models
presented. Conversion of data series to stationary sequences proves to be an inevitable
process to carry out time-series analysis efficiently and the second chapter of results is
delving into this issue. Finally, this work is also enhanced by a short analysis and
description of a «shifting» problem we encountered during different evaluation
3
processes and experiments, with a short explanation and a path to avoid it
Collections
Show Collections