![Page 1: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/1.jpg)
DALEXDescriptive mAchine Learning EXplanations
Alicja GosiewskaMI2 Data Lab
Warsaw University of Technology
![Page 2: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/2.jpg)
Data and models
![Page 3: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/3.jpg)
Data and models
![Page 4: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/4.jpg)
Data and models
![Page 5: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/5.jpg)
The explain() function
explain(model, data, y, predict_function, link, ..., label)
![Page 6: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/6.jpg)
Model performance
![Page 7: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/7.jpg)
Model performance
![Page 8: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/8.jpg)
auditor: model performance
![Page 9: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/9.jpg)
Variable importance
![Page 10: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/10.jpg)
Variable response
![Page 11: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/11.jpg)
Improving the model
- Linear model and random forest had equal performance for apartments dataset.
- In general the random forest model has smaller residuals than the linear model but there is a small
fraction of very large residuals.
- Random forest model under-predicts expensive apartments. It is not a model that we would like to
employ.
![Page 12: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/12.jpg)
Improving the model
- Linear model and random forest had equal performance for apartments dataset.
- In general the random forest model has smaller residuals than the linear model but there is a small
fraction of very large residuals.
- Random forest model under-predicts expensive apartments. It is not a model that we would like to
employ.
- `construction_year` is important for the random forest model.
- the relation between `construction_year` and the price of square meter is non linear.
![Page 13: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/13.jpg)
Improving the model
![Page 14: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/14.jpg)
Improving the model
![Page 15: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/15.jpg)
![Page 16: Descriptive mAchine Learning EXplanations · Improving the model - Linear model and random forest had equal performance for apartments dataset. - In general the random forest model](https://reader033.vdocument.in/reader033/viewer/2022041703/5e433d2ffc5c26445a7a4402/html5/thumbnails/16.jpg)
We acknowledge the financial support from the NCN Opus grant
2016/21/B/ST6/02176
Acknowledgements