中文摘要：

提出了一种基于决策树C4.5的多示例学习算法C4.5-MI,通过拓展C4.5的熵函数和信息增益比来适应多示例学习框架.应用梯度提升方法对C4.5-MI算法进行优化,得到效果更优的GDBT-MI算法.与同类决策树算法在benchmark数据集上进行比较,结果表明,C4.5-MI和GDBT-MI算法具有更好的多示例分类效果.

英文摘要：

A new multi-instance learning algorithm C4. 5-MI based on decision tree C4. 5 was proposed,and the entropy function and information gain ratio to multiple instance framework were extended. Excellent algorithm GDBT-MI was obtained by improving C4. 5-MI through adopting gradient boosting. The results on several benchmark datasets demonstrated the effectiveness of C4. 5-MI and GDBT-MI over other similar algorithms.