Predicting Market Values Using Machine Learning

Motivacija iza ovog projekta jest pitanje može li osobi koja ulazi u vode trgovanja pomoći umjetna inteligencija s modelima koje nudi. Za ovaj projekt, odabran je programski jezik Python kao jedan od najsuperiornijih programskih jezika za umjetnu inteligenciju i strojno učenje. Zadatak rada je usporediti dva algoritma strojnog učenja, Long Short-Term Memory i Višeslojni Perceptron, i zaključiti koji je bolji za predviđanje buduće cijene valutnog para odnosno dionice na američkom tržištu. Koristeći podatke s Foreign Exchange i New York Stock exchange tržišta, na temelju povijesnih podataka o cijeni dionica, algoritmi imaju zadaću prepoznati trend u kojoj se neki tržišni par nalazi te predvidjeti buduće cijene. Na temelju usporedbe korijena srednje kvadratne pogreške dvaju algoritama, možemo zaključiti kako je Long Short- Term Memory algoritam puno bolje predvidio buduće cijene na američkom tržištu.Motivation behind this project is the question if a model in the Artificial Intelligence space can help people, who want to start their trading carrers, make better decisions. For this project, programming language Python was used as one of the most superior programming languages for Artificial Intelligence and Machine Learning. The task of the paper is to compare two machine learning algorithms, Long Short- Term Memory and Multilayer Perceptron, and to conclude which one is better for predicting future prices of currency pairs or shares on the American market. Using data from the Foreign Exchange and New York Stock Exchange markets, based on historical stock price data, the algorithms have the task of recognizing the trend in which a market pair is located and predicting future prices. Based on the comparison of the root mean square error of the two algorithms, we can conclude that the Long Short-Term Memory algorithm predicted future prices of the American financial assets much better