自2013年Aileen Lee提出独角兽企业概念以来，社会广泛关注这一重要的新兴群体。独角兽企业一方面体现了“创新”，另一方面体现了“高成长”。中国独角兽企业也处于快速发展阶段，企业数量和企业估值总额全球排名第二，仅次于美国。本文通过描述性统计对比分析中美两国独角兽企业的现状，发现与美国相比，中国独角兽企业胜在“高估值”而非“数量多”，“高估值”主要体现在头部少数企业，且独角兽企业保持估值高速增长的难度较大。本文从中国独角兽企业最新估值和估值增长速度两个角度研究估值，旨在得到区分企业估值高低、估值增长速度快慢的重要影响因素及其影响路径，为企业、投资者、政府提供一定的建议，从而促进中国独角兽企业的发展。

本文将《2023年胡润全球独角兽之中国企业榜单》中的企业作为研究样本，综合文献与数据可得性，最终构建了包含企业基本特征、企业创新能力、企业所属区位、企业融资情况和企业所属行业五个维度的指标体系，运用梯度提升树（GBDT）算法进行建模分析，得到各因素的相对重要性及重要因素的影响路径，最后对比分析不同机器学习方法的优劣。主要结论：（1）GBDT模型拟合效果好。最新估值模型在测试集上AUC为0.88，估值增长速度模型在测试集上AUC为0.83。（2）对于最新估值模型，影响企业是否为“高估值”的14个变量中，相对重要性超过10%的有2个变量，超过5%的有7个变量。（3）对于估值增长速度模型，影响企业是否为“高成长性”的14个变量中，相对重要性超过10%的有4个变量，超过5%的有8个变量。（4）对于最新估值模型，相对重要性最大的三个变量是最新融资金额、注册资本和社保人数。（5）对于估值增长速度模型，相对重要性最大的三个变量是最新融资金额、最新融资时间距今的月数和专利数。（6）对比分析Logit、支持向量机和随机森林算法的模型拟合效果，发现GBDT是最佳模型。

Since Aileen Lee introduced the concept of unicorns in 2013, there has been widespread interest in this important emerging group. Unicorns embody "innovation" on the one hand and "high growth" on the other. China's unicorn enterprises are also in a rapid development stage, ranking second in the world in terms of number of enterprises and total enterprise valuation, only after the United States. This paper analyzes the current situation of unicorn enterprises in China and the United States through descriptive statistics comparison, and finds that compared with the United States, China's unicorn enterprises win in "high valuation" rather than "high number", and "high valuation" is mainly reflected in the head of a small number of unicorn enterprises, and it is more difficult for unicorn enterprises to maintain high valuation growth. This paper studies valuation from the perspective of the latest valuation and valuation growth rate of China's unicorn enterprises, aiming to get the important influencing factors and their influencing paths that distinguish the valuation of enterprises from the valuation of enterprises, and the valuation growth rate from the valuation of unicorn enterprises, so as to provide enterprises, investors, and the government with certain suggestions, and thus to promote the development of Chinese unicorn enterprises.

This paper takes the enterprises in the "2023 Hurun Global Unicorn List of Chinese Enterprises" as the research samples, integrates the literature and data availability, and finally constructs an index system containing five dimensions: enterprise basic characteristics, enterprise innovation ability, enterprise location, enterprise financing and enterprise industry, and applies the Gradient Boosting Decision Tree (GBDT) algorithm to conduct the modeling analysis to get the relative importance of each factor and the influence paths of important factors, and finally compare and analyze the advantages and disadvantages of different machine learning methods. Main findings: (1) The GBDT model fits well. The AUC of the latest valuation model is 0.88 and the AUC of the valuation growth rate model is 0.83. (2) For the latest valuation model, among the 14 variables affecting whether a company is "highly valued" or not, the relative importance of 2 variables exceeds 10%, and the relative importance of 7 variables exceeds 5%. (3) For the valuation growth rate model, of the 14 variables affecting whether a company is "high growth", 4 variables have a relative importance of more than 10% and 8 variables have a relative importance of more than 5%. (4) For the latest valuation model, the three variables with the highest relative importance are the latest amount of financing, registered capital and number of people in social security. (5) For the valuation growth rate model, the three variables with the greatest relative importance are the latest financing amount, number of months since the latest financing and number of patents. (6) Comparatively analyzing the model fitting effects of Logit, Support Vector Machine and Random Forest algorithms, GBDT is found to be the best model.