课堂教学评测是了解教学水平与提升教学质量的重要依据。传统的课堂评测主要通过问卷与量表的方式进行，但这种方式耗时耗力，且不可避免地会引入主观误差，降低了评测结果的准确性与可信度。近年来，智慧教育的兴起与发展提供了更为便捷与高效的现代化教育形式，也为课堂教学评测带来了新的思路。在此背景下，本课题以加速教育信息化和智慧教育建设为宗旨，构建出一套基于统计建模和集成学习的课堂教学评测系统，具体工作如下：

首先调研并提炼出一套适用于传统量表与AI分析相结合的课堂教学评测指标体系；之后提出基于层次分析法与熵权法的主客观融合统计评测模型与基于AdaBoost的集成学习评测模型，并根据统计模型与集成学习模型在不同课堂教学评测指标上的性能互补性，构建出综合评测模型；最后针对不同评测指标，设计该系统的可视化方案，分别生成学生课堂学习评测报告与教师课堂教学评测报告，并在web端进行可视化呈现。经过实验测试，本课题的综合模型针对分类指标评测的准确率普遍在80%-90%之间，针对回归指标评测的均方根误差均在10左右，性能表现较好。

In-class teaching evaluation is a significant process for comprehending teaching level and improving teaching quality. Currently, the common method of in-class teaching evaluation is keeping track of teaching and learning in courses and analyzing related questionnaires and scales from education experts, but it relies heavily on experts’ subjectivity thus brings inevitable subjective error and cannot reach a high accuracy, which remains an important challenge in setting an effective and reliable evaluation method. In recent years, Smart Education has attracted extensive interest because of its remarkable great effect by combining information technology and education. Thereby, this article propose a statistical modelling and ensemble learning-based system for in-class teaching evaluation with the aim of accelerating the construction of Smart Education in teaching evaluation. Firstly, a set of in-class teaching evaluation indexes suitable for the combination of traditional scales and artificial intelligence assessment are refined. Secondly, a comprehensive evaluation model is proposed, which combines the “analytic hierarchy process-entropy weight method” statistical model and the AdaBoost-ensemble learning model. Lastly, teachers’ in-class teaching evaluation report and students’ in-class learning evaluation report are automatically generated and visualized on the web. The proposed system is used in current in-class teaching data set to test its performance. The comprehensive evaluation model achieves the accuracy of 80%-90% in classification indexes and the Root Mean Square Error of approximate 10 in regression indexes. We suggest that this system based on statistical modeling and ensemble learning promotes a new in-class teaching evaluation pattern.