<p>Predicting stock prices through machine learning has emerged as a prevalent approach in recent years and many researchers utilize forecasting results to construct trading strategies. Although prior research has used machine learning to predict stock price movement, they only consider a single fixed trading strategy in the trading process. As market conditions evolve, such a single fixed strategy may decline in effectiveness and struggle to sustain outstanding performance in all periods. This paper introduces a methodology for constructing a self-adaptive trading strategy based on predicting stock prices using machine learning. This self-adaptive strategy dynamically adjusts to market changes, ensuring its competitiveness over all periods. Through empirical experiments on the Hong Kong stock market, it can be found that this method can indeed yield better performance. This method achieves the highest returns and effectively controls the risk.</p>