We use bottom-up approach in individual options to predict the trend of the U.S. semiconductor stock market in the next four weeks. The ten-year rolling ranking of the volatility smirk and the P/(C+S) ratio, and the four-month ranking difference of the first two are taken as model variables, and we use Random Forest and XGBoost algorithms to predict the crash signal of the stock price index. The strategy result shows that the model has a good performance when predicting index collapsing more than 6.37%; in addition, the difference of the volatility smirk ranking in the short term can improve the model’s ability to predict crash signal, while the ranking difference of the P/(C+S) ratio in the short term needs other variables to distinguish between useful information and noise.