This study explores the effectiveness of machine learning and eye movement features in predicting Chinese reading proficiency. Unlike previous research, which focused on one or two specific levels of eye movement features, this study integrates passage-, sentence- and word-level eye movement features to predict reading proficiency. By analyzing the eye movements of 71 native Chinese-speaking undergraduate students as they read nine short passages, a support vector machine was constructed to predict Chinese reading proficiency. Proficiency was determined based on performance on the Chinese achievement test in the National College Entrance Examination and scores from the cloze test. The results indicate that the model, which utilizes passage-, sentence- and word-level eye movement features comprehensively, achieves the highest prediction accuracy (81.69%, 84.71%). Nevertheless, eye movement features at the word, sentence, and passage levels each play a unique role in predicting Chinese reading proficiency. The results provide empirical support for the relationship between eye movement features at different levels and the reading proficiency of Chinese readers. The outcomes highlight the feasibility of integrating eye movement features at the passage, sentence, and word levels, and of employing support vector machine to construct a predictive model for the reading proficiency of Chinese readers.