A combination of machine learning based natural language processing with technical analysis for stock trading

Abstract

Stock price analysis appropriately is a challenging area of research as many factors directly affect stock prices. As a result, so not easy to analyze to identify stock trading signals appropriately. The proposed approach builds a framework for classifying stock trading signals by combining natural language processing with technical analysis. The dataset implemented focuses on corporate news and stock indicators from 01-01-2019 to 31-12-2021 from the eight corporates of the Thai Industry Group Index and Sector Index. Two traditional machine learning models, multilayer perceptron (MLP) and support vector machine (SVM), and four deep learning models, Bidirectional GRU (BiGRU), bidirectional LSTM (BiLSTM), gated recurrent unit (GRU), and long short-term memory (LSTM) used for comparison purposes. The training model classifies daily trading signals into three classes: buy, sell, and hold-after that, the model’s efficiency evaluates by measuring accuracy, precision, recall, and F1-score. For the results, classification average efficiency in all models showed that the BiGRU model obtained higher average accuracy (0.93), precision (0.93), recall (0.93), and F1-score (0.92) than other models. Therefore, the BiGRU model was appropriate for our experiment and was applied to determine daily trading signals for analyzing investment returns.