服务国家战略创新统计赋能——基于多行业人工智能劳动力替代风险的统计建模与赋能研究

Serving National Strategy through Innovative Statistics and Empowerment—Statistical Modeling and Empowerment Research on AI-Driven Labor Substitution Risks across Multiple Industries

摘要: 在数字经济与人工智能广泛渗透背景下，精准评估多行业人工智能劳动力替代风险，对产业转型、就业稳定和政策制定意义重大。研究遵循“数据预处理–EDA–模型构建–验证优化–聚类分析–结论建议”流程。数据预处理采用四步特征筛选：以Pearson相关系数和VIF剔除共线性，结合卡方检验与随机森林重要性交叉验证，最终保留6个核心特征，规避数据泄露与筛选漏洞。模型构建设分类(监管壁垒等级)与回归(自动化风险指数)双目标，选用GBR、GBC等，经调参与融合，确定GBR + GBC加权融合为最优。采用SHAP进行全局和局部解释，发现替代弹性是核心特征，影响方向符合经济逻辑。通过轮廓系数、Calinski-Harabasz指数和类内方差显著性检验验证K-means聚类有效性，刻画不同风险等级行业特征。以替换样本、替换算法、剔除高影响行业三项检验验证稳健性，结果显示模型稳定、泛化能力强。研究发现：制造、零售、物流等劳动密集型及部分服务密集型行业为人工智能替代高风险领域；创意、教育等依赖人类创意与情感交互的行业为低风险领域；不同行业替代风险的核心影响因素差异显著。

Abstract: Against the backdrop of the widespread penetration of the digital economy and artificial intelligence, accurately assessing AI-driven labor substitution risks across multiple industries is of great significance for industrial transformation, employment stability and policy formulation. This research follows the workflow of “data preprocessing - EDA - model construction - validation and optimization - cluster analysis - conclusions and recommendations”. A four-step feature selection procedure is adopted in data preprocessing: Pearson correlation coefficient and VIF are used to eliminate multicollinearity, combined with cross-validation via Chi-square test and random forest feature importance. Six core features are retained in the end to avoid data leakage and selection bias. The model construction sets dual objectives including classification (regulatory barrier level) and regression (automation risk index). Algorithms such as GBR and GBC are adopted. After parameter tuning and ensemble learning, the weighted ensemble of GBR + GBC is identified as the optimal solution. SHAP is applied for global and local interpretation. The results show substitution elasticity serves as the core feature, whose impact direction conforms to economic logic. The validity of K-means clustering is verified using the silhouette coefficient, Calinski-Harabasz index and significance test of intra-cluster variance to characterize industrial features at different risk levels. Robustness is tested through three approaches: sample replacement, algorithm replacement and removal of highly influential industries. The results demonstrate that the model is stable with strong generalization ability. The study finds that labor-intensive industries including manufacturing, retail and logistics, as well as some service-intensive sectors, are high-risk areas of AI substitution; industries relying on human creativity and emotional interaction such as creativity and education are low-risk areas; core influencing factors of substitution risk vary markedly across industries.