462
O. Ogorodnyk et al.
for Knowledge Analysis 2019). 10-folds cross-validation was used to estimate the
skill of the model on the new data. The correlation coefficient, mean absolute error
and root mean squared error was used to estimate the difference between the measured
and predicted values of width and thickness of the dogbone parts.
32.4.1 REPTree (Decision Tree)
REPTree stands for Reduced Error Pruning Tree. It is a decision tree algorithm that
follows the regression tree logics through the creation of multiple trees in different
iterations (Kalmegh 2015), information gain is used as a splitting criterion in this
case. The best tree from all the generated ones is then chosen and considered as
representative. Pruning of the tree is done using the mean square error on the predictions of the tree. The result of the algorithm application is a predictive model in the
form of a decision tree or a set of if–then rules. In this study minimum number of
instances in a leaf has been tested to see how it would influence the prediction quality
of the obtained model.
32.4.2 Random Forest
Random forest is an ensemble learning algorithm that includes many separate learners
(Breiman 2001). Each tree is created using a random sample of cases from the dataset.
The collection of several tree predictors, in this case, is called a forest. In the case
of regression, the model’s output “is the average of the responses over all the trees
in the forest” (Kalmegh 2015). In the experiment, different numbers of trees in the
forest are tested.
32.4.3 K-Nearest Neighbors
The k-nearest neighbors algorithm uses parameter similarity to predict values of
the new data points (Kononenko and Kukar 2007). The new point is assigned value
depending on the values of instances closest to it. The number of closest instances is
the k input parameter of the algorithm. In the case of regression, the average value
of the closest data points is taken as the model’s output. Euclidean distance is used
in the experiment to calculate distances between the data points, a different number
of nearest neighbors is also used to build the models.
O. Ogorodnyk et al.
for Knowledge Analysis 2019). 10-folds cross-validation was used to estimate the
skill of the model on the new data. The correlation coefficient, mean absolute error
and root mean squared error was used to estimate the difference between the measured
and predicted values of width and thickness of the dogbone parts.
32.4.1 REPTree (Decision Tree)
REPTree stands for Reduced Error Pruning Tree. It is a decision tree algorithm that
follows the regression tree logics through the creation of multiple trees in different
iterations (Kalmegh 2015), information gain is used as a splitting criterion in this
case. The best tree from all the generated ones is then chosen and considered as
representative. Pruning of the tree is done using the mean square error on the predictions of the tree. The result of the algorithm application is a predictive model in the
form of a decision tree or a set of if–then rules. In this study minimum number of
instances in a leaf has been tested to see how it would influence the prediction quality
of the obtained model.
32.4.2 Random Forest
Random forest is an ensemble learning algorithm that includes many separate learners
(Breiman 2001). Each tree is created using a random sample of cases from the dataset.
The collection of several tree predictors, in this case, is called a forest. In the case
of regression, the model’s output “is the average of the responses over all the trees
in the forest” (Kalmegh 2015). In the experiment, different numbers of trees in the
forest are tested.
32.4.3 K-Nearest Neighbors
The k-nearest neighbors algorithm uses parameter similarity to predict values of
the new data points (Kononenko and Kukar 2007). The new point is assigned value
depending on the values of instances closest to it. The number of closest instances is
the k input parameter of the algorithm. In the case of regression, the average value
of the closest data points is taken as the model’s output. Euclidean distance is used
in the experiment to calculate distances between the data points, a different number
of nearest neighbors is also used to build the models.