本章概要
本章讲述了模型评估与选择(model evaluation and selection)的相关知识:
2.1 经验误差与过拟合(empirical error & overfitting)
精度accuracy、训练误差(经验误差)training error(empirical error)、泛化误差**generalization error、过拟合**overfitting、欠拟合underfitting;
2.2 模型评估方法(evaluate method)
测试误差testing error、留出法hold-out、分层采样stratified sampling、交叉验证法cross validation、k-折交叉验证**k-fold cross validation、留一法leave-one-out(LOO)、自助法bootstrapping、自助采样bootstrap sampling、包外估计out-of-bag estimate、调参**parameter tuning、验证集validation set;
2.3 模型性能度量(performance measure)
错误率error rate、查准率(准确率)precision、查全率(召回率)recall、P-R曲线、平衡点BEP、F1/Fβ、混淆矩阵、ROC曲线、AUC、代价敏感cost-sensitive、**代价矩阵**cost matrix、代价曲线cost curve、期望总体代价;
2.4 模型比较检验(comparation & testing)
假设检验hypothesis test、拒绝假设、t-检验t-test、Friedman检验、后续检验post-hoc test、Friedman检验图;
2.5 偏差与方差(bias & variance)
偏差-方差窘境bias-variance dilemma;