In this repository an adaboost classifier based on linear SVM was designed to classify the dataset combining 'classA.csv' and 'classB.csv'. Class A and Class B datasets are visualized below in Figure 1.
Figure 1. Datasets 'classA.csv' and 'classB.csv'.
A linear SVM with varying C values {0.1, 1 , 10 , 100} was trained and the decision boundary was visualized alongside the reported accruuacy based on 10-times-10-fold cross validation.The linear SVM decision boundary is shown in Figure 2 for varying C values.
Figure 2.Decision boundary of varying C values.
The accuracy results of the linear SVM classifiers obtained using 10-times-10-fold cross validation are reported in Table I.
TABLE I: LINEAR SVM ACCURACY RESULTS
| C Value | 0.1 | 1 | 10 | 100 |
|---|---|---|---|---|
| Accuracy - Mean | 79.35% | 79.80% | 79.77% | 79.75% |
An ensemble of classifiers was created based on the Adaboost-M1 approach to classify the dataset. A linear SVM was used with the C=1 as the weak learner classifier. The maximum number of weak learners was taken as T = 50.
The ensemble model decision boundary using Linear SVM is shown in Figure 3.
Figure 3. Decision boundary of ensemble model.
The mean and variance of the accuracy results for the 10-times-10 fold cross validation of the ensemble learning using linear SVM are reported in Table II.
TABLE II: LINEAR SVM ENSEMBLE LEARNING RESULTS
| Accuracy | Variance | Standard Deviation |
|---|---|---|
| 89.29% | 23.98% | 4.90% |