机会型创业对跨越中等收入陷阱，实现经济转型升级和高质量发展的作用日益凸显。基于计划行为理论，从主观规范、行为态度、知觉行为控制3个方面并结合人口统计学特征选择12个特征变量，构建机会型创业影响因素框架。在此基础上，基于全球创业观察数据库，运用XGBoost算法预测机会型创业并判别关键影响因素，将预测结果与3种机器学习算法进行比较。结果表明，基于准确率、精确率、召回率和F1值4个评估指标，XGBoost算法可以较好地预测机会型创业，优于逻辑回归、支持向量机和随机森林算法；创业自我效能、机会识别和关系感知是影响机会型创业的重要因素。聚焦新型创业研究，有助于拓展计划行为理论的适用边界和机器学习算法在创业领域的应用，为有效识别机会型创业和针对性培育机会型创业提供理论指导与实践启示。

Opportunity-driven entrepreneurship plays an increasingly prominent role in overcoming the middle-income trap, and realizing economic transformation and high-quality development, especially in the face of an increasingly complex international environment and the impact of the global COVID-19 pandemic. Existing studies have pointed out some influencing factors of opportunity-driven entrepreneurship, but they are relatively fragmented. As an influential behavioral prediction theory in the field of social psychology, the theory of planned behavior can effectively predict entrepreneurial intention and subsequent entrepreneurial behaviors. It provides a comprehensive and powerful explanatory framework for opportunity-driven entrepreneurship. At the same time, entrepreneurial activity is a complex social issue and is subject to the dynamic nature of nonlinear network feedback system. As a result, it is hard to predict entrepreneurial activity. The significance level of traditional regression methods is influenced by the sample size and the coefficient is influenced by the measurement scale, which makes it difficult to assess the importance of influencing factors as well. Opportunity-driven entrepreneurship involves plenty of characteristic variables, and the relationship between each variable and opportunity-driven entrepreneurship may not be limited to a single linear relationship.#br#With the advent of the era of big data and the development of computer information technology, artificial intelligence receives increasing attention in more and more research and application. It is disrupting industries, business management and innovation related to entrepreneurship, but little attention is paid to the field of entrepreneurship. As one of the core algorithms in the field of artificial intelligence, the machine learning algorithm can fit nonlinear relationships well and is suitable for dealing with the prediction of opportunity-driven entrepreneurship. Therefore following the theory of planned behavior, the study selects 12 characteristic variables to construct the influencing factor framework of opportunity-driven entrepreneurship from subject norm, attitudes toward the behavior and perceived behavior control combined with demographic characteristics. The influencing factor framework specifically includes achievement orientation, risk taking, public recognition, media publicity, entrepreneurial self-efficacy, etc. A total of 19 805 samples are obtained from the Global Entrepreneurship Monitor database in 2018. The XGBoost algorithm is used to predict opportunity-driven entrepreneurship and identify key influencing factors, and the results are compared with those of three machine learning algorithms. The results show that the XGBoost algorithm can predict opportunity-driven entrepreneurship better than support vector machine, random forest and logistic regression in accuracy, precision, recall and F1 value. According to the ROC curve, the AUC value can reach 0.94, which again shows that the prediction effect of XGBoost algorithm is good. In the meantime, entrepreneurial self-efficacy, opportunity recognition and relationship perception are important factors affecting opportunity-driven entrepreneurship, while demographic characteristics such as gender are less important.#br#This study makes two contributions.It first applies the theory of planned behavior and demographic characteristics to analyze the influencing factors of opportunity-driven entrepreneurship, which helps make up for the lack of systematicity in the analysis of opportunity-driven entrepreneurship in previous studies, verifies the applicability of planned behavior theory in explaining opportunity-driven entrepreneurial behavior, and deepens the application of the theory in the field of opportunity-driven entrepreneurship. Then the study combines artificial intelligence with opportunity-driven entrepreneurship, applies machine learning methods to the prediction of opportunity-driven entrepreneurship, and establishes a prediction model of “whether individuals participate in opportunity-driven entrepreneurship”. The XGBoost algorithm has the best prediction effect on opportunity-driven entrepreneurship, and can detect the interaction ambiguity and nonlinear effects in the input data, extend the application of machine learning methods in the field of entrepreneurship, and make up for the shortcomings of traditional econometric analysis methods.#br#The study also provides practical insights and references for individuals, government and external investors. First, for government and external investors, they can identify potential opportunity-driven entrepreneurs based on the XGBoost algorithm to facilitate better allocation of resources to potential opportunity-driven entrepreneurs and reduce selection costs to promote opportunity-driven entrepreneurial practice. Second, the government should target to create an environment that fosters opportunity-driven entrepreneurship to continuously improve individual entrepreneurial self-efficacy, enhance individual entrepreneurial relationship perception, and strength individual entrepreneurial opportunity recognition. For individuals, special attention should be paid to enhancing entrepreneurial self-efficacy, focusing on the accumulation of entrepreneurial relationships and actively identifying entrepreneurial opportunities to transform into opportunity-driven entrepreneurial behavior.#br#