Trends in Digital Signal Processing

May 27, 2015 12:43 PSP Book - 9in x 6in 17-Yong-Ching-Lim-c17

Conclusion 529

great care must be taken when using it on large applications since

it is required to store all the data internally to build the model.

SVM is another classifier that performed well in all our

experiments. Like kNN, it requires some user supplied parameters.

Unlike kNN, it is likely to scale up to bigger problem sizes because

it keeps a memory efficient internal model. However, understanding

theory behind SVM may not be a trivial task.

The final algorithm we tested was the Decision Tree classifier.

Due to over-fitting problems, its performance was significantly

lower than all the other ...