# Interpretations of this residual value scatterplot of LinearRegression GridSearch CV model

I am very new to machine learning and did a first mini project predicting house price using GridSearchCV, elastic net model.

Then I plotted out the residual value (Y_Predict - Y_Train) but I want to know whether the graph tells my model is good or bad? How do I translate it into valuable write-up? Or is there any other graph that I can use to tell more stories? I am sort of clueness about what else I can do based on this model, as this is my first model and I dont really know about the options so far. Thank you in advance!

P.S. The MSE I got is 14195, RMSE is 20558, mean value of the dataset label "SalePrice" is 180815.

• "Good" or "bad" depends on the intended application. Suppose your client were a person wishing to fix a selling price for their house. How willing do you suppose they would be to pay you for your estimate if you represented--as is consistent with this plot--that about half the time you can estimate the correct price to within 10% or so?
– whuber
Nov 8, 2021 at 19:05

Okay, The thing about residual plot is, If you find any patterns forming, It indicates a problem in your model. There is no specific pattern forming in the residual plot given.

Moreover a Mean Absolute error of 119 is not at all bad for this data set. That means on an average, Your prediction are off by 119. This may not be enough but this is a good indicator to show that you are proceeding in the right direction.

You can do on more thing, If this is a 2 feature data-set, You can plot out the Test values actual prediction graph vs the true test values and see the smoothness of the line its fitting

• There is a fairly clear upward trend in the residuals at higher prices. There are several low outliers. Mean absolute error is probably the wrong measure of residuals: for house prices, in most applications, relative error would be more relevant. BTW, where do you obtain "119" as a mean absolute error??
– whuber
Nov 9, 2021 at 15:13
• Thank you!Could you please advise what i can do to improve the "upward trend for higher prices' residuals" @whuber Also, you mentioned for house prices relative error is more suitable - may I ask how to procure such domain knowledges? Any learning path and resource you can please recommend? Nov 10, 2021 at 9:24
• Domain knowledge is obtained through study and by collaborating with experts. (With housing prices, some common sense is involved. For instance, somebody looking at houses in the \\$100K range will see no value in an estimate that could err by that much, immediately telling us to consider relative errors.) To deal with the upward trend, you might consider a model that can handle nonlinear relationships. These are many and varied, ranging from including a quadratic term through splines through transforming one or more of the variables and on into more exotic procedures.
– whuber
Nov 10, 2021 at 11:40