XLSTAT: Extreme Gradient Boosting (XGBOOST) in Excel

 

Dataset for Example

The dataset is extracted from the data science platform, Kaggle, and might be accessed at this address.

The “Banknotes” dataset comprises a list of 200 banknotes with some information: This dataset contains 6 variables; One is qualitative and concerns the banknotes' authenticity and the others are quantitative and related to banknotes' shapes.

Counterfeit: in case the banknote is genuine we put “0”, in the contrary case the banknote is counterfeit we put “1”. 100 banknotes are counterfeits when the 100 others are genuine in this dataset. Length, Left, Right, Bottom, Top are quantitative variables.

 

Setting Up

• Open XLSTAT.

• Click on the Machine learning menu and select the Extreme Gradient Boosting feature.

 

In the General tab:

• In the Response variable field, select the "Counterfeit" variable.

• Set the Response type to binary because the response variable has only two distinct values.

• In the Quantitative explanatory variables, select all the remaining variables in the dataset.

• In the Options tab, many parameters are available to set up the model. Here we let the default parameters.

• In the Validation tab, select to keep 30 random observations for testing model performance.

• In the Outputs tab, enable the confusion matrix, Variable Importance (set to Gain), and Results by Objects.

• Click OK to run the computations.

Results will appear on a new Excel sheet.

 

Interpreting the Results

The misclassification rates show model performance: 1.2% for training and 10% for validation.
The confusion matrix displays classification accuracy per class—98.78% for genuine (modality 0) and 98.86% for counterfeit (modality 1).
The confusion plot visually compares observed (gray) and predicted (orange) counts, showing nearly perfect overlap with 81/82 and 87/88 correct predictions for modalities 0 and 1, respectively.

 

The Results by Object table shows each observation’s true class, predicted class, and probabilities for each response category.

 

The Variable Importance table shows which features best predict banknote authenticity.
The Gain metric measures each feature’s relative contribution—higher Gain means greater importance.
Here, BOTTOM (bottom margin width) is the most important feature, suggesting a link to authenticity.

 

Conclusion

In this example, 97.65% of observations are correctly classified, with the lower margin being the key feature for authenticity. The 90% accuracy on validation data confirms good model generalization.