I made a Support Vector Machine in Python using the sklearn library. The data I am using is Amplitude vs Frequency for 200-520 kHz with a 0.33 kHz step. My input data is the amplitude value for my n-data samples at each of the 961 frequencies I have recorded, making it a n by 961 matrix. I have two classification groups 1 and 2. I have included an image of a sample of what my spectra data looks like below.
When I change the Kernel I use in the model, between Linear, Polynomial, Radial Basis Function and Sigmoid, only the Linear Kernel gives me a Recall, Precision and F-Measure value above 0.50, all other Kernel's are give me values less than 0.3, and mostly predict the entire dataset to be either group 1 or group 2. Does this make sense with regards to spectra data that a linear Kernel would perform the best? I am very new to AI and Support Vector Machines, however I could not find anything online about this.
Any comments/ feedback would greatly be appreciated.
Recall = (True Positive)/(True Positive + False Negative)
Precision = (True Positive)/(True Positive + False Positive)
F-Measure = (2*Recall*Precision)/(Recall + Precision)