Abstract
This study presents a comparative evaluation of traditional machine learning algorithms and state-of-the-art large language models (LLMs) in classification tasks using structured data. The analysis is conducted on the widely used Iris dataset, which consists of measurements from three different flower species and serves as a benchmark in machine learning literature. The performance of classical algorithms, including logistic regression, K-nearest neighbors, decision trees, and random forest, is examined based on well-established evaluation metrics. Following the implementation of these algorithms in Python (Python Software Foundation, Wilmington, DE, USA), their results are compared with the numerical reasoning capabilities of LLMs, including GPT-5.3 Instant (OpenAI, San Francisco, California, USA), Gemini 3.0 Flash (Google LLC, Mountain View, California, USA), Copilot (Microsoft Corporation, Redmond, Washington, USA), and Qwen 3.5 (Alibaba Group, Hangzhou, China). The LLMbased evaluations are performed by providing algorithm-specific prompts and analyzing the generated outputs. The findings indicate that, beyond their natural language processing capabilities, LLMs can achieve competitive performance in structured data classification tasks, reaching accuracy levels between 90% and 97%. Among the evaluated models, Gemini 3.0 Flash demonstrates superior performance by more effectively capturing multivariate relationships within the dataset. Overall, the results suggest that transformer-based architectures can serve as strong alternatives to traditional deterministic algorithms in numerical classification problems, highlighting the evolving role of LLMs as general-purpose data analysis tools.