Explaining machine learning models for age classification in human gait analysis
Explaining machine learning models for age classification in human gait analysis
复制标题
解释人类步态分析中年龄分类的机器学习模型
DOI：
10.1016/j.gaitpost.2022.07.153
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
B. Horsak
中科院分区：
文献类型：
--
作者：
D. Slijepcevic;Fabian Horst;Marvin Simak;Sebastian Lapuschkin;Anna;W. Samek;C. Breiteneder;W. Schöllhorn;M. Zeppelzauer;B. Horsak
方法我们利用了AIST步态数据库2019[6]的一个子集，其中包含健康参与者赤足行走时每人五次双侧地面反作用力(GRF)记录(图1A)。每个输入信号在拼接前进行最小-最大归一化，然后送入卷积神经网络(CNN)。受试者分为三个年龄组：青年(20-39岁)、中年(40-岁)和老年(65-79岁)。结果平均分类准确率为60.1±4.9%，明显高于零规则基线(37.3%)。混淆矩阵(图1B)显示，CNN很好地区分了年轻人和老年人，但很难对中年人进行建模。LRP显示，对于年轻人，最相关的区域是GRF_AP和GRF_V的第二峰；对于中年人，GRF_ML和GRF_V的第一峰和倾向于第二峰的区域最相关。对于老年人，讨论根据LRP，所有GRF信号中都存在与年龄分类相关的区域。文献[7]、[8]、[9]支持GRF_AP和GRF_V第二峰的相关区域，而GRF_V第一峰的相关区域不支持。某些相关区域，如GRF_V的第一峰(老年人)和第二峰(中年人)的倾向性，以及GRF_ML的区域，过去没有调查过，并为未来的研究提出了问题。我们的结果表明
MethodsWe utilized a subset of the AIST Gait Database 2019 [6] containing five bilateral ground reaction force (GRF) recordings per person during barefoot walking of healthy participants (Figure 1A). Each input signal was min-max normalized before concatenation and fed into a Convolutional Neural Network (CNN). Participants were divided into three age groups: young (20–39 years), middle-aged (40–64 years), and older (65–79 years) adults. The classification accuracy and relevance scores (derived usingResultsThe mean classification accuracy of 60.1±4.9% was clearly higher than the zero-rule baseline (37.3%). The confusion matrix (Figure 1B) shows that the CNN distinguished younger and older adults well, but had difficulty modeling the middle-aged adults. LRP showed that for young adults, the most relevant regions were the second peak in GRF_AP and GRF_V. For middle-aged adults, regions in GRF_ML and the first peak and incline to the second peak of GRF_V were most relevant. For older adults, theDiscussionAccording to LRP, relevant regions for age classification reside in all GRF signals. Relevant regions at the second peak of GRF_AP and GRF_V are supported by the literature [[7],[8],[9]], whereas the relevant region at the first peak of GRF_V is not. Certain relevant regions, eg, the incline to the first (older adults) and second (middle-aged adults) peak in GRF_V, as well as regions in GRF_ML, were not investigated in the past and raise questions for future research. Our results suggest
影响因子：
2.4
作者：
Halilaj E;Rajagopal A;Fiterau M;Hicks JL;Hastie TJ;Delp SL
通讯作者：
Delp SL