Financial markets are fundamental to the stable development of the socioeconomic landscape. This study focuses on the efficient market hypothesis (EMH) positing that financial markets reflect all available information. The importance of studying EMH lies in assessing its feasibility in contemporary society and trying to make markets become "efficient". This paper reviews the theoretical models of EMH, explores statistical validation methods, and investigates leveraging machine learning for stock price prediction. Those findings suggest that while the feasibility of EMH in today's society may be reduced, its theoretical framework and derivative formulas still hold reference value. Furthermore, when utilizing machine learning for stock price prediction, it is essential to consider subjective human factors, such as irrational behavior. In conclusion, this research highlights the diminished feasibility of EMH in modern society, underscores the continued improvements of its theory and formulas, and emphasizes the significance of considering subjective human factors in utilizing machine learning for stock price prediction.