As the plot,HDD_OSA range is from 30 to 100,but the predicted OSA range is from 65 to 75.
How to understand this ? how to improve the predicted range?
Please help,thanks!
JMP Ver:16.1
Hi
预测模型的问题是你的预测数据分布集中在65-75. 而且模型的拟合效果并不理想。当你的预测结果集中在65-75，但是实际结果的数据范围是35-95。因此残差太大，R平方就很低。如果方便的话是否可以提供一下数据源？否则不太好说具体是什么问题。
The plots suggest that your model is poor. The is a weak association between the predictors and the response, so most of the prediction is the mean response.
Hi,Frank
原始数据如下，请参考：
响应:HDD_OSA
训练参数: Inner_EW_R2,OW_R2,QST_AMP, QST_STO_R, MRR2
预测的结果确实不理想，我的疑问是：R2是不好，但为什么预测值的范围被限定在65到75？激活函数的限制吗?谢谢！
Yes,the model is poor ,I pay more attention to the predicted range.
Try use Python,change the activation function from tanh to relu, seems the predicted range become better.
Not sure the predicted range is too meaningful here, as indicated per earlier comments. You might also try Model Screening if you have JMP Pro. Also, some simple univariate analysis (in Distribution) reveals that there may be some resolution limitation in the HDD_OSA response:
Here I generated a model that predicts similarly (low R-squared for training and validation sets) but for which the SSE and -Loglikelihood of the Validation set are lower that the initially proposed model:
Also probably good to be aware of the higher correlation between Inner_EW_R2 and OW_R2.
You may benchmark this one against your Python model with RELU.
