模型预测效果评价,通常用相对绝对误差、平均绝对误差、根均方差、相对平方根误差等指标来衡量
一般回归型模型评价
绝对误差Absolute Error
相对误差Relative Error
平均相对误差Mean Absolute Error【MAE】
均方误差Mean Squared Error【MSE】
均方根误差Root Mean Squared Error 【RMSE】
一般分类型模型评价
- TP(True Positives)是指正确的肯定表示正确肯定的分类数;
- TN(True Negatives):正确的否定表示正确否定的分类数;
- FP(False Positives):错误的肯定表示错误肯定的分类数;
- FN(False Negatives):错误的否定表示错误否定的分类数。
准确率 【Accuracy】
精确率 【Precision】
提取出的正确信息条数 / 提取出的信息条数
召回率 【Recall】
提取出的正确信息条数 / 样本中的信息条数
混淆矩阵【Confusion Matrix】
F1得分 【F1-Score】
F1_Score=正确率∗召回率∗2/(正确率+召回率)F_1\_Score=正确率 * 召回率 * 2 / (正确率 + 召回率) F1_Score=正确率∗召回率∗2/(正确率+召回率)
ROC曲线
PR曲线
library(ROCR)
library(gplots)
train_predict <- predict(lda.model, newdata &#61; traindata)
test_predict <- predict(lda.model, newdata &#61; testdata) par(mfrow &#61; c(1, 2))
predi <- prediction(train_predict$posterior[, 2], traindata$MN)
perfor <- performance(predi, "tpr", "fpr")
plot(perfor, col &#61; "red", type &#61; "l", main &#61; "ROC曲线", lty &#61; 1)
predi2 <- prediction(test_predict$posterior[, 2], testdata$MN)
perfor2 <- performance(predi2, "tpr", "fpr")
par(new &#61; T)
plot(perfor2, col &#61; "blue", type &#61; "l", pch &#61; 2, lty &#61; 2)
abline(0, 1)
legend("bottomright", legend &#61; c("训练集", "测试集"), bty &#61; "n", lty &#61; c(1, 2), col &#61; c("red", "blue"))
perfor <- performance(predi, "prec", "rec")
plot(perfor, col &#61; "red", type &#61; "l", main &#61; "PR曲线", xlim &#61; c(0, 1), ylim &#61; c(0, 1), lty &#61; 1)
perfor2 <- performance(predi2, "prec", "rec")
par(new &#61; T)
plot(perfor2, col &#61; "blue", type &#61; "l", pch &#61; 2, xlim &#61; c(0, 1), ylim &#61; c(0, 1), lty &#61; 2)
abline(1, -1)
legend("bottomleft", legend &#61; c("训练集", "测试集"), bty &#61; "n", lty &#61; c(1, 2), col &#61; c("red", "blue"))