The distribution of stock market capitalization in the A-share market usually has obvious right-skewed and fat-tailed characteristics, which means that a few large stocks account for most of the total market capitalization, while the vast majority of stocks have a small market capitalization. This distribution characteristic will cause large stocks to have too much impact on the factor value when the market capitalization factor is used directly, which may distort the prediction effect of the model. The logarithmic transformation compresses the original market capitalization data into a smaller range, making the factor distribution closer to the normal distribution. This transformation reduces the impact of extreme values, helps the model pay more attention to the overall market characteristics, and improves the universality of the model on stocks of different market capitalizations. At the same time, the logarithmic transformation can also reduce the heteroscedasticity in the data, which helps to improve the effectiveness of statistical inference.