构建基于小波熵的自训练半监督支持向量机分类模型评价老年人步态

The Self-Training Semi-Supervised Support Vector Machine Based on Wavelet Entropy for the Evaluation of the Elderly Gait

研究应用半监督学习算法分析未标注步态数据评价老年人步态,提出基于小波熵的自训练半监督支持向量机步态分类模型,通过小波熵从未标注步态数据中选取为每次自训练步态分类模型所需最具信息量的标注样本,有效获取步态数据类别间和步态数据内在的"有价值"的步态变异信息,提高步态分类器的泛化性能。首先采用10名老年人和10名青年人步态数据构建支持向量机分类模型,然后对120名不同年龄组未标注步态数据分类预测,依据小波熵选取样本数据,逐步添加更新步态样本训练集,自训练支持向量机分类模型。实验结果表明,本算法较准确鉴别青年和老年人步态模式(分类正确率90%),比基于有监督学习的支持向量机步态分类算法正确率提高近5%,有效改善支持向量机步态分类算法性能,有望为临床提供一个评价老年人步态的新方法。

This work investigated the application of the semi-supervised learning algorithm to analysis of the unlabeled gait data for evaluating the elderly gait. The novel gait classification algorithm that the self-training semi-supervised support vector machine （SVM） based on wavelet entropy for the discrimination between the young and elderly gait pattern was addressed. In the self-training, wavelet entropy was employed to obtain labeled samples from unlabeled dataset. The most valuable information related to the gait change was acquired for current gait classifier model, which obviously improved the gait classification performance of SVM. The labeled gait sample datasets including 10 young and 10 elderly participants were used to develop SVM that was employed to classify the unlabelled gait dataset from 120 subjects of different age groups. The new labeled gait data, obtained by our defined wavelet entropy, were selected and constructed the new sample train dataset for developing self-training SVM. The experimental results showed that the accuracy of our proposed algorithm is 90% in recognization of the young and elderly gait pattern. Furthermore, the accuracy of our proposed algorithm was increased approximately 5% compared with that of the classification algorithm by the supervised support vector machine, suggesting that our proposed technique can obtain more information related to gait change from the labeled and unlabeled gait dataset, and provides a new tool for assessment of elder gait.