Stock price prediction has long been an attractive issue for stock investors since it has the potential to maximize trading profit. Traditionally, stock investors usually tried to predict future stock prices or trends by experience-based intuition or simple mathematical modeling approaches, e.g. technical indicators. With the progress of machine learning, new methods have been proposed to tackle the stock price prediction problem. This paper presents a novel E3s-CNN-BiLSTM model based on advanced neural networks for stock price prediction. To evaluate the performance of the proposed model, experiments based on historical stock market data from TWSE were conducted to compare it with previous methods. Experimental results show that our E3s-CNN-BiLSTM model achieves the best performance among all methods. The experimental results also reveal that different input data constituents have significant impact on a stock price prediction model’s performance, and thus need to be paid attention.