The study aims to enhance real-time prediction of e-commerce data trends through the design and evaluation of an integrated deep learning model optimized using Keras Tuner. The primary objective is to develop an efficient Keras Tune-based LSTM model capable of accurately capturing and predicting dynamic customer behaviour patterns in e-commerce platforms. To assess its effectiveness, the optimized model is compared against baseline models such as standalone CNN, LSTM, and CNN-LSTM with attention mechanisms. The pro- posed framework leverages Keras Tuner for automated hyperparameter optimization, ensuring improved performance and adaptability. Experimental results on real-world e-commerce datasets demonstrate a significant improvement in predictive accuracy and model efficiency, validating the superiority of the optimized model in supporting real-time business intelligence and decision-making.