the aim is to show the most important features contributing to target variable.
Assalamu aleikum!
1. The conditions of the problem to be solved;
2. The structure of the data set;
3. The general theory of the method;
4. The optimal parameters;
5. The results obtained.
Dear Fereshteh Ehsani
Please write the algorithm and mathematics behind the formation of decision tree on your dataset. It will create a huge impression on the reviewers that you are able to understand the actual logic of creation of tree. Also method which type of decision tree algorithm you are going to use, it may be either CART, ID3, C4.5, CHAID or MARS etc. These things you can put into the research methodology section. For the results you can put the optimal tree drawn from your algorithm and accuracy obtained, misclassification rate, precision and recall. You can also give the confusion matrix.
Hope you find this useful.
Thanks,
Gyanendra Chaubey
Dear Fereshteh Ehsani ,
you should first report the decision tree (DT) model parameters (e.g. maximum number of decision splits, the minimum number of leaf node samples, split criterion);
then you have to report the optimization procedure you used to optimize the model, including validation type (k-fold cross-validation, leave-one-out validation, held-out validation) and objective function (e.g. F-score, accuracy);
you should also report a comprehensive performance evaluation, including recall/sensitivity, specificity, precision/positive predictive value, negative predictive value, accuracy, F-score. You better include also the receiver operating characteristic (ROC) curve and respective area under the curve (AUC) value, highlighting your working point on the curve.
finally, you can report a picture of the model structure, to make it easy to understand the most important features (i.e. that close to the root of the tree) and the process of learning (going from the root to the leaves of the tree, one can figure out which is the decision process for each sample).
Gyanendra Chaubey
thank you it was really helpful.
if i have used sklearn for my DT, should i mention the basic method of this algoritm?
due to large numbers of explanatory variables and as a result the big size of the obtained tree, can i just draw first 5 rows of it and drop the rest?
Luigi Borzí
thank you... this answer helped me alot.
could you suggest any article or format which contains these sections?
due to large numbers of explanatory variables and as a result the big size of the obtained tree, can i just draw first 5 rows of it and drop the rest rows?
Dear Fereshteh Ehsani
I hope that before implementing the model you have the EDA and Feature selection part. If you have done that well consider your hyperparameters and draw the tree. If still it is taking more than one page to draw the tree then you can leave it and insert the confusion matrix, classification report etc. to show your model accuracy.
Thanks
Gyanendra
Tree structure
Depth of Tree
Maximum number of terminal nodes
Minimum samples at a leaf node
Minimum Samples for a node split
Maximum features
Computational time
Accuracy
confusion matrix using k-fold cross-validation
Precision, Recall, and F1 Score
Kappa stats
- Badges
- Science topic