CODEWITHSUNDEEP
machine-learningdecision-tree
setCodesToFire
setCodesToFire

Reputation: 49

Decision Trees performance

When does a decision tree perform well. I plotted some graphs comparing a model based on decision tree and another using logistic regression. Decision tree took longer time to build the model while the LRclassifier took less time.Moreover the f-score of logistic is more than the decision tree for this model. So I want to know when should decision tree be used.

Upvotes: 1

Views: 408

Answers (3)

Stephen Flynn
Stephen Flynn

Reputation: 125

A single decision tree run somewhere around 75%-80% accuracy when the training data is relevant. "Decision trees by random forest" improves accuracy further. That is usually when there are several decision trees trained and they "vote" on an outcome. Decision trees are typically used to find the most relevant dimensions to train neural networks from. Neural networks have greater capacity to discover hidden correlations in data.

Upvotes: 0

Aditya
Aditya

Reputation: 1058

The questions looks to a bit incomplete. You should have also shown those plotted graphs and also the details of the data on which you have applied both these models.

But still Decision tress are highly interpretable as compared to linear and logistic regression. In simple terms DT are just nested if-else statements. By interpretbility i mean your model should give the reason for why it has chosen a specific class as a label and not just say that the query point belongs to class 0 or 1.

Upvotes: 0

indigo
indigo

Reputation: 94

Usually you generate the model as a one-off upfront and then use it to classify new instances of data. So you don't have to worry too much about how long it takes to generate the tree unless you are having to generate it dynamically with real-time data or some such scenario.

Upvotes: 1

Related Questions