Files
Dawance_33861600_2024.pdf
Closed access - Adobe PDF
- 1.43 MB
Details
- Supervisors
- Degree label
- Abstract
- During this master thesis will be explores the behaviour and capabilities of the DL8.5 decision tree algorithm for classification and regression tasks with several new functions. More specifically firstly will be analysed the evolution of the accuracy during a cross-validation on several datasets when modifying several parameters of the DL8.5 function on classification tasks. There will be a comparison of the results with another decision tree algorithm, CART and also with different other simple classification models. Also a logistic regression will be performed in the leaves of the tree. Secondly there will be a focus on regression tasks. Regression is computed with two different approaches of DL8.5, the Cluster function and the Predictor function. During the regressions the criteron of interest is the evolution of the RMSE in comparison to diverse other existing models. Lastly will be presented the implementation of model trees. This method consists of training a function inside each leaf of the tree and then using this function during the prediction of new values. This method allows more complex models on data than decision trees or models independently.