Forecasting time series with deep learning and artificial intellgence

In this study, analyzing the S&P data obtained from the Yahoo Finance website and making future predictions as a result of the analysis have an important place for computer science and many investors. Thanks to the developing technologies, the power of the computers has increased and by this means more powerful algorithms have been developed. It has become possible to make predictions on financial data using these algorithms. In the present study, it was aimed to predict financial data via deep learning and shallow learning and daily data of the S&P 500 index between 12.08.2000 and 13.8.2020 were used in the study. In this context, the first 19 years of the data (0.95 per cent of the data) were used in the training steps, and the only last 1 year of data was put to use as estimation (test data). Long short-term memory and multilayer perceptrons methods were used for estimation. During the examination of the obtained results, it was observed that the training and test errors of the multilayer perceptron and the long short-term memory methods were close to each other. Furthermore, while training, test and all-data root mean square errors are calculated as 17.3, 65.3 and 22.0 dollars respectively for the long short-term memory network; it was found to be 16.1, 61.2 and 20.6 dollars for the multilayer perceptron network.