Predicting Thai stock index trend using deep neural network based on technical indicators

Abstract

In this study, we aimed to find a suitable model for predicting the direction of the Stock Exchange of Thailand index (SET50 index) by developing a deep neural network model that builds upon the advancements of a hybrid model of an artificial neural network and genetic algorithm. Due to the complexity of stock data and the challenging predictability, a single hidden layer may not be sufficient. Therefore, we proposed a deep neural network model with three hidden layers, optimizing the number of nodes in each layer to achieve accurate predictions of the movement of the index. The input data consists of technical indicators widely used by technical stock analysts. These indicators are calculated over four different lookback periods of 3, 5, 10, and 15 days. The data was collected from the SETSMART system, which can retrieve historical data and real-time data via an API. We focused on data from the period of 2015–2019, comprising 1,220 records. Our test results showed that the proposed model achieved the highest average accuracy at 82.94%, outperforming the previous model.