I have worked on individual several supervised classifiers using weka.
Can I combine two or more classifiers. How can it be done in matlab
or weka?
First you have to calculate the classification rate of each classifier, then you have several methods to achieve the combination. A simple method is weighted mean. To do that you can use this equation: sum(si * tj) / sum(tj) where si represent the output of each classifier and tj the classification. From a threshold value you will decide the optimal deision. This is the combination of decisions of classifiers for an optimal decision.
This is a complicated question to answer. I suggest you look into the field of ensemble learning, which is concerned with combining individual classifiers to form better decisions.
There are numerous possible solutions.
As previously mentioned, a weighted mean is a simple solution (you can have a classifier instead of the weights). Or a simple agreement method would also be possible.
I would also suggest you to have a look at semi-supervised theory. Co-Training might be an interesting solution to optimize the training of your models of both your classifiers.
Finally you can also have a look at Late Fusion solutions which proposes several solution for merging classifiers output. (This topic is often adressed in mutlimodal contexts). I would suggest you to read a bit about late fusion mechanisms
you may want to have a look at this article for a general overview of fusion
Niaz, U.; Merialdo, B., "Fusion methods for multi-modal indexing of web data," Image Analysis for Multimedia Interactive Services (WIAMIS), 2013 14th International Workshop on , vol., no., pp.1,4, 3-5 July 2013
doi: 10.1109/WIAMIS.2013.6616129
And this one for a more details on specific weights in late fusion
Lai, K.; Liu, D.; Chen, M.; Chang, S., "Learning Sample Specific Weights for Late Fusion," Image Processing, IEEE Transactions on , vol.PP, no.99, pp.1,1
doi: 10.1109/TIP.2015.2423560
Hope this helps
I think there are two ways:
1. Ensemble learning.
2. There is a field related with rank list combination. I think it can be usd to classifier combinations. For example:
James Allan, Anton Leuski, Russell Swan, Donald Byrd, Information processing and management, evaluating combinations of ranked lists and visualizations of inter-document similarity, 2001
Good luck.
Thank u so much for your valuable suggestion. Your answers given me a open eye
for my work.
Ensemble Learning.. And simply "VOTE"ing ensemble.. You can use as much as classifiers on top of different "voting schemes".
I think, you should start with sample datasets from UCI web-link. And make some experiments to see the results quantitatively. Then read some articles to absorbe the underlying theory.
WEKA also supports many other ensemble algorithms with a user friendly GUI.
good luck
- Badges
- Science topic