Performance Augmentation of Base Classifiers Using Adaptive Boosting Framework for Medical Datasets
Table 2
Results for precision, F-measure, and recall error of all the classifiers in AdaBoost framework for various data sets.
Classifiers
Data sets
Leukaemia
Lymphoma-I
Lymphoma-II
GCM
Data set C
Instances
38
45
96
144
60
Attributes
7130
4027
4027
16064
7130
Naïve Bayes
Recall
0.947
0.911
0.75
0.167
0.6
F-measure
0.946
0.911
0.692
0.048
0.56
Precision
0.951
0.914
0.683
0.028
0.552
Voted perceptron
Recall
0.789
0.978
xx
0.167
0.6
F-measure
0.799
0.978
xx
0.048
0.58
Precision
0.847
0.979
xx
0.028
0.583
Stacking
Recall
0.711
0.511
0.479
0.167
0.65
F-measure
0.59
0.511
0.31
0.048
0.512
Precision
0.505
0.511
0.23
0.028
0.423
Adaboost
Recall
0.895
0.867
0.51
0.167
0.583
F-measure
0.895
0.867
0.445
0.094
0.52
Precision
0.895
0.87
0.403
0.066
0.5
Bagging
Recall
0.921
0.933
0.865
xx
0.633
F-measure
0.92
0.933
0.84
xx
0.629
Precision
0.92
0.934
0.836
xx
0.626
J48
Recall
0.842
0.822
0.835
xx
0.567
F-measure
0.842
0.822
0.835
xx
0.546
Precision
0.842
0.825
0.864
xx
0.536
Random tree
Recall
0.789
0.644
0.667
0.382
0.683
F-measure
0.789
0.643
0.649
0.373
0.685
Precision
0.789
0.645
0.648
0.37
0.687
Random forest
Recall
0.791
0.778
0.781
0.521
0.65
F-measure
0.761
0.776
0.743
0.509
0.643
Precision
0.791
0.788
0.75
0.53
0.639
Bayes network
Recall
0.933
0.978
xx
0.167
0.633
F-measure
0.933
0.978
xx
0.048
0.629
Precision
0.934
0.979
xx
0.028
0.626
Decision stump
Recall
0.895
0.867
0.51
0.167
0.633
F-measure
0.895
0.867
0.445
0.094
0.629
Precision
0.895
0.87
0.403
0.066
0.626
Zero-R
Recall
0.711
0.444
0.479
0.167
0.65
F-measure
0.59
0.429
0.31
0.048
0.512
Precision
0.505
0.434
0.23
0.028
0.423
Input mapped classifier
Recall
0.711
0.444
0.479
0.167
0.567
F-measure
0.59
0.429
0.31
0.048
0.571
Precision
0.505
0.434
0.23
0.028
0.576
The crossed (xx) cells show that the results could not be generated for the specific classifier because of the limitation of the framework or the data set. Hence, the evaluation of these classifiers’ results has been carried out manually to check if any better results could be gathered for comparison. The results in bold indicate the best results for different datasets.
