Ensemble Learning from Imbalanced Data Set for Video Event Detection

Learning from imbalanced data sets is a hot and challenging research topic with many real world applications. Many studies have been conducted on integrating sampling-based techniques and ensemble learning for imbalanced data sets. However, most existing sampling methods suffer from the problems of information loss, over-fitting, and additional bias. Moreover, there is no single model that can be applied to all scenarios. Therefore, a positive enhanced ensemble learning (PEEL) framework is presented in this paper for effective video event detection. The proposed PEEL framework involves a novel sampling technique combined with an ensemble learning mechanism built upon the base learning algorithm (BLA). Exploratory experiments have been conducted to evaluate the related parameters and performance comparisons. The experimental results demonstrate the effectiveness of the proposed PEEL framework for video event detection.