Normalization of the FFT histograms

Dr. Bunn suggested normalizing FFT histograms for the SVM to be able to generalize better. This seemed logical and we tried it out. Surprisingly, the results were not better, infact, for quake predictions, the results were more likely to be predicted correctly by a random guess (less than 50% accuracy). But, the number of false positives, did go down considerably.

Results:
Training Data:
Even the training data was not fit by the SVM trained on it.
True positive: 48.90%

True Negative: 97.23%

False Positive: 2.77%

False Negative: 51.10%

Test Data:
Since even the training data was not fit by the SVM, we did not even expect the test data results to look good.
True positive: 30%

True Negative: 100%

False Positive: 0%

False Negative: 70%