A Survey on Automated Data Analysis Techniques Powered by Large Language Models

This survey provides a detailed examination of automated data analysis techniques empowered by Large Language Models (LLMs). We first categorize core data analysis tasks across descriptive, diagnostic, predictive, and prescriptive stages, highlighting how LLMs enable natural language interfaces for querying, code generation, and workflow synthesis. Emphasis is placed on LLM-driven automation in data preparation, exploratory analysis, insight generation, and interactive visualization, illustrating the growing integration of LLMs as orchestrators of complex analytical processes. We discuss advances in leveraging LLMs for multi-step reasoning, sub-task decomposition, and domain-specific language-guided workflow construction, which collectively enhance interpretability and modularity. Despite notable progress, challenges remain in addressing hallucination, domain adaptation, and seamless system integration. This survey synthesizes recent research trends and practical frameworks, providing a foundation for future developments in building robust, explainable, and usercentric LLM-based data analysis systems.