Adaptive stock price volatility forecasting using enhanced kMedoids clustering with attention based LSTM for day trading

Forecasting price changes for stocks is complicated because they are impacted by numerous factors and because market phenomena are extremely volatile. In this paper, we propose a new predictive modelling approach combining an advanced clustering k Medoids where the number of clusters is determined automatically by the Silhouette criterion and an attention-enhanced Long short term memory (LSTM) network. This approach increases the predictive power of stock price movements. Stock prediction models classifies stocks with similar patterns and attention-enhanced LSTM or LSTMs focuses on the critical logs. This method is tested on eleven stocks listed on the BSE and NSE using of intraday and historical datasets. Out of all tested models, the proposed ek Medoids—attention LSTM model reduced the NRMSE (Normalized Root Mean Squared Error) and NMAE (Normalized Mean Absolute Error) by 9.4% and 7.8% when compared to the baseline LSTM model and ek Medoids—LSTM models. Although we have shown improvements in prediction accuracy of stock prices, in the interest of producing actionable stock trading signals, the study examines the model’s ability to produce real-time BUY, SELL and HOLD (BSH) trading signals using live stock market data. In summary, we conclude that the predictive modelling approach offers stock traders a reliable stock price prediction tool and an actionable trading system for short-horizon forecasting and intraday decision-making.