Sectoral Stock Prediction Using Convolutional Neural Networks with Candlestick Patterns as input Images

Abstract

Stocks were one of the instruments for investment that could provide benefits rather than just saving money, but investments could also result in losses if wrong in investing in shares. Therefore determining the time frame beforehand will make easier to predict the trend of stock movements within a certain time period. Stock prediction analysis method which was quite popular among traders until now was to used candlesticks. Candlestick patterns were often used by stock prediction analysis in several centuries ago. The form of candlestick could infer the direction of prices. CNN, RNN and LTSM were currently widely used for forecasting in stock prediction movement. The most suitable method for image datasets was CNN. In this paper, we presented CNN and candlestick approach to recognize an image to identify the strength of a trend pattern in the prediction movement of stock. Based on our experiment, CNN with candlestick approach can produce up to 99.3% accuracy overall. Therefore, this method can produce good accuracy.