Comparing the classification success rate of modulation schemes using SVM at different SNR values, it is found that the success rate is slightly lowered by 0.6% when comparing 20dB SNR with 10dB. What can be a logical explanation for this slight decrease when the SNR is increased from 10 to 20dB.
It is not strange, It may varies with the change input values, however., you may follow some standard procedures as other researchers follows. I suggest you follow : Conference Paper SVM-based classification of digital modulation signals
https://arxiv.org/pdf/1901.05850.pdf
Dear Mohamed
One possible explanation for the slight drop of about 0.6% in performance when you increased the SNR from 10dB to 20dB into your dataset could be because the 20dB SNR would make the dataset more random in nature, thus distorting the regularity within the dataset. And this would in turn make it difficult for the SVM to classify such pattern compared to when you introduced a 10dB SNR into the same dataset. In other words, the higher the SNR introduced into the dataset the more random/irregular the inherent patterns in the dataset set would be, thus resulting in lower classification outcomes on the part of the SVM classifier.
I had a similar experience in my previous works when I introduced White Gaussian Noise (WGN) of different SNR into my dataset. I found increasing the SNR of the WGN would generally lead to a corresponding decrease in accuracy.
I hope you would find these tips helpful.
Thank you for the valuable answers Dr. Muhammad Ali and Dr. @Oluwarotimi Williams Samuel
- Badges
- Science topic
- Similar topics
- Data Mining
- Classification