Utility of converting time series data into an image for classification purposes

Duboko učenje je jedno od područja računalstva koje je postalo sve popularnije i razvijenije posljednjih godina i to je područje koje se promatra unutar ovog rada. Točni aspekti koji se promatraju je usporedba dvije arhitekture neuronskih mreža. Promatramo rekurentnu i konvolucijsku neuronsku mrežu (RNN i CNN) s naglaskom na konvolucisjku neuronsku mrežu u smislu da koristimo podatke koji nisu originalno namijenjeni za tu arhitekturu te hoće li ta mreža s prilagodbom podataka postići zadovoljavajuće rezultate koji su usporedivi s arhitekturom i podatcima korištenim za rekurentnu neuronsku mrežu. Korišteni podatci su snimke moždanih valova dobivene koristeći EEG namijenjeni za rad s RNN arhitekturom, te se prilagodba podatka za CNN arhitekturu izvodila koristeći GAF (GASF i GADF) i MTF algoritme za generiranje slika iz vremenskih podataka.Deep learning is one of the fields of computer science that has become ever more popular and advanced in recent years and that is the field which is observed within this paper. Exect aspect that is observed is the comparison of two neural network architectures. We observe the recurrent and the convolutional neural network (RNN and CNN) while concentrating on the convolutional neural network in the sense that we use data that is not originally suitable for that architecture and will that network with the adjusted data give a satisfying results that are comparable with the architecture and data used for a recurrent neural network. The used data are recordings of brain waves that are acquired by using an EEG and are intended for RNN architecture but for the CNN architecture the data was adjusted by using GAF (GASF and GADF) and MTF algorithms for generating images from time series data