3.2output.txt

The following data was generated using the Support Vector Machine classifier.

--------------------------------------------------------------------
| Number of Samples per Class  |  % of Test Set Correctly Classified 
--------------------------------------------------------------------
|    500                                              52.6462 %    |
|   1000                                              54.5961 %    |
|   1500                                              54.039  %    |
|   2000                                              54.8747 %    |
|   2500                                              54.039  %    |
|   3000                                              53.4819 %    |
|   3500                                              53.7604 %    |
|   4000                                              54.3175 %    |
|   4500                                              53.2033 %    |
|   5000                                              53.4819 %    |
|   5500                                              52.9248 %    |
--------------------------------------------------------------------

Here we can see that the best classification rate was achieved when
using 2000 samples of each class. As we increase the number of 
samples, accuracy tends to decrease. A notable exception is for 
4000 samples, where we observe a classification rate of 54.3175, 
the third best result from our testing. This may or may not be
explained by random sampling noise.

The general decrease in accuracy can potentially be explained by the classifier 
overfitting to the training data. As we train on more data points, 
we lose the capacity to generalize well on unseen data.