32 Prediction of Width and Thickness of Injection Molded …
465
Fig. 32.3 Obtained structure of a decision tree for prediction of thickness1
Since pruning was enabled during the algorithm’s work, not all the 25 parameters
were included in the tree. The decision tree uses holding pressure, plasticizing time,
pressure at switchover, injection time, average cushion value, injection work, backpressure, cushion after holding pressure and injection speed as parameters to take
the decision about the value of the thickness of the dogbone specimen.
For example, if looking at the tree from the very top, we can easily get the
following: if the value of the holding pressure is less than 27.5, and value of plasticizing time is less than 1.53, then, on average, the value of the width of the first
dogbone will be 3.39. Numbers in the brackets indicate the number of samples
belonging to that leave and number of instances from the pruning.
32.5.2 Discussion
The results show that all the machine learning methods used in this study show the
high capability of predicting the width and thickness of the focus parts. The best
results in terms of the correlation coefficient are achieved using the random forest
algorithm, while the second-best result is received through the application of the
MLP. At the same time, REPTree decision tree performs a little bit better, than kNN.
However, overall all four methods show quite high prediction capabilities on this
data. Overfitting should not be present since the 10-folds cross-validation technique
has been used in order to estimate skill of the models on the new data.
465
Fig. 32.3 Obtained structure of a decision tree for prediction of thickness1
Since pruning was enabled during the algorithm’s work, not all the 25 parameters
were included in the tree. The decision tree uses holding pressure, plasticizing time,
pressure at switchover, injection time, average cushion value, injection work, backpressure, cushion after holding pressure and injection speed as parameters to take
the decision about the value of the thickness of the dogbone specimen.
For example, if looking at the tree from the very top, we can easily get the
following: if the value of the holding pressure is less than 27.5, and value of plasticizing time is less than 1.53, then, on average, the value of the width of the first
dogbone will be 3.39. Numbers in the brackets indicate the number of samples
belonging to that leave and number of instances from the pruning.
32.5.2 Discussion
The results show that all the machine learning methods used in this study show the
high capability of predicting the width and thickness of the focus parts. The best
results in terms of the correlation coefficient are achieved using the random forest
algorithm, while the second-best result is received through the application of the
MLP. At the same time, REPTree decision tree performs a little bit better, than kNN.
However, overall all four methods show quite high prediction capabilities on this
data. Overfitting should not be present since the 10-folds cross-validation technique
has been used in order to estimate skill of the models on the new data.