A Low-Code Data Analytics Agent with Flexible Functional Module Selection for Employment Data Mining

Abstract

To address the practical challenges of employment data analysis — the complexity of the data that makes manual analysis difficult, the inefficiency of manually written scripts, and the hallucination problem of large-language-model (LLM) data analysis — this work builds on my earlier EPQ Project using a static random forest modeling approach and investigates an employment data analysis agent on a low-code agent building platform. We propose an integrated agent framework comprising four modules: data loading and cleaning, statistical data analysis, machine learning modeling, and data visualization. We design a workflow orchestration prompt that drives the large model to flexibly invoke and combine the four modules on demand, enabling fully automated employment data mining. Two experiments are conducted, and the results are compared with those of our earlier EPQ Project, which verifies that the agent delivers prediction accuracy, implementation efficiency, application flexibility, and broad applicability and enables end-to-end automation. Applicable to any structured employment dataset, the agent offers high analysis efficiency and strong prediction accuracy, effectively resolving the difficulties of manual analysis of complex structured employment data and the problem of weak prediction accuracy.