Figure 4

Parameter Filtering Curve. Variation of accuracy with number of parameters used in SVM training for the Overall, Animal and Plant models. To investigate whether a smaller set of parameters were sufficient for training, the SVM was successively retrained for a growing set of parameters that was incremented by one for each training cycle. Beyond 25 parameters, there was little improvement in accuracy, and when more than 30 parameters were used, the accuracy began to drop with significantly less accuracy achieved when all 44 parameters were used (far right of plot). See Materials and Methods for full details. Arrows indicate the number of parameters selected for each model.