How to improve svm model accuracy using weka?
Hi All,
I have using svm classification for my dataset.But roc values is 0.65.How to improve my svm model accuracy? anybody suggesting me.
• 4,600 views
•
link
written
by
krishavc92 •
0 answers
No answers yet.
Log in to answer this question.
More posts like this
-
Improve predictive models and validation based on ROC
written by n@4j •
-
How should I draw a roc curve to generalize the performance of a model?
written by ssko •
-
Is the significance of p-value between different ROC curves of machine learning models important?
written by egascon •
-
What is the use of the Kraken2 nt database?
written by Natalia •
-
how to get feature importance after SVM
written by pt.taklifi
-
hap.py roc curve plot
written by Nicolas Rosewick
-
How do i set different threshold to get multiple values for ROC plot
written by Paul
-
Colorectal cancer stage multiclass-classification using Gene Expression Data
written by victorya •
-
how to parallelize single function in R?
written by na.cna30 •
-
Svm-Predict Input File Format
written by Panos
Without knowing more about your data and what you're doing it's hard to give you good advice. In general, the first thing that would improve accuracy of a classifier would be a good/better training set. The representation of the data item also has some importance i.e. the features you use and/or the kernel. Assuming the data contains the information needed to infer the classes, you could try to figure out where the poor performance comes from i.e. high bias vs high variance (see here for more on this). Finally, consider that maybe this accuracy reflects what's in the data. Ask yourself whether given the data you or an expert would do better.
Thanks for reply i have four feature for my data. In this data training set having 1429 instance and test set 158 instance. I run supplied test option for using weka and i calculated cost and gamma values for training set data using libsvm tool then i put proper cost and gamma values whatever i get libsvm tool.can you suggest me which kernal type is best for libsvm classifier.
How many classes do you have ? If you have just a few classes and 4 features, try some exploratory analysis first to get an idea of whether your features contain enough information about the classes e.g. look at correlation between features and classes, try clustering to see if the classes are already separable in the original feature space. Which kernel to use depends on the data. The idea is to find one that make the classes linearly separable in kernel feature space. Often people use an RBF kernel by default. Note that you need to tune the parameters.
I have two classes and five features.