A comparative study of deep learning model with binary and multiple classification to predict stock market trends by detecting fractal patterns based on Elliott wave theory.

One of the popular but complicated methods in technical analysis is the Elliott wave method. In this method, the most important part is to recognize the main trend patterns of the market, which is a difficult task due to the fractal structure of the market. But like other fields, the use of artificial intelligence in the field of financial forecasts has also become widespread. Therefore, it seems that the use of artificial intelligence in Elliott wave analysis is attractive. Therefore, in the current research, by introducing a deep learning model to predict the market through the detection of Elliott wave patterns, it has been investigated and compared the power of the model in two modes of binary and multiple classification. In this research, for 15 considered patterns, 1002 examples of stock price charts of companies present in the Iranian stock market in the 11-year period from 1390 to 1400 were collected and labeled, and finally for recognition as input to the deep learning algorithm with Recurrent neural network model was used in binary and multiple classification modes. In this research, RapidMiner 9.9 software was used to design and implement the model, and accuracy criteria were used to determine the power of the model. The results show 18% accuracy in pattern recognition in multiple classification mode and 61% accuracy in binary classification mode. Therefore, the power of the deep learning model in detecting the fractal patterns of Elliott waves and as a result predicting the market trend is significantly higher in the binary classification mode than in the multiple classification mode. Therefore, the present study recommends the use of deep learning model with binary classification to detect fractal patterns of Elliott waves.