A real challenge related to the efficient market hypothesis (EMH) is that past information will not work in forecasting because it reflects the price. Some classic methods are often used to rely on statistical and quantitative models. However, forecasting becomes more difficult when variables in the model are non-stationary, and relationships between variables sometimes are very weak or occur simultaneously. Ho Chi Minh city stock market (HOSE) is an inefficient emerging market. Besides, constant development with powerful features of algorithms in machine learning and deep learning has opened a promising new direction. The objective of this study is to compare the forecasting capacity of three predictive models: Support Vector Machine (SVM), Long Short-Term Memory Recurrent Neural Network (RNN), and Logistic Regression (LR). The data used is the code of shares in the VN30 list from 1/01/2015 to 27/01/2022, and trading activities take place at the end of each day. This study used the" rolling window," resulting in the RNN forecast to excel at an average of 82.19%. In addition, Logistic Regression plays an important role in interpreting the statistical significance of input variables. From the experimental results, this study proposes that researchers use machine learning algorithms to ensure the accuracy of the prediction. At the same time, investors should pay attention to company characteristics when making investment plans in the medium and long term.