Stock market prediction is challenging due to its complex behaviour. Various theories are available for stock trend prediction. Technical analysis is widely used by researchers in which various charts and statistics are used to analyse the market trend. Candlestick charts are a useful tool in technical analysis. However, utilizing candlestick charts with machine learning techniques is still a matter of research. This paper proposes a model which uses candlestick chart patterns for the trend prediction of Indian and US stock indices. At first, candlestick patterns are extracted from the historical stock data which results in the generation of a large number of features. In this work, the Gaussian-modified PSO (g-PSO) is proposed for feature selection and an optimized K-Nearest Neighbour (KNN) model for trend prediction. The proposed model is tested on the four stock indices BSE, NIFTY50, S&P500, and DJIA. The performance metric Accuracy, Precision, Recall, and F1-score are used for the model assessment. The proposed model gives an average accuracy of 61.4%. The proposed model is compared with the stateof-the art methods and comparatively provides better results

Stock Market Forecasting, Candlestick Chart, Machine Learning (ML), K-nearest Neighbour (KNN), Particle Swarm Optimization (PSO), Grid Search