Are there any classification algorithms implementations in R such as Decision Trees, Naive Bayes, etc. in which the training instances can have a weight?
I found this relevant question in CrossValidated but it seems it is for regression. I am looking for something similar for classification problems.
this might be considered a hack but - in a neural net classifier (such as a multi-layer perceptron trained through backpropagation) the weight of an individual example could be manipulated by including many copies of that example in the training set
the neural net package is the goto in R and there are many nice tutorials out there which would work for the purpose of testing the effect of varying the proportion of the repeated example to give it extra weighting
As a little wilder of a suggestion (although not in r) - in cognitive science there is also a concept known as selective attention that can be applied to a classification/category learning problem. check out ALCOVE (kruschke, 1992) or SUSTAIN (love et al., 2004) - a more modern approach if these seem interesting
refs
Kruschke, J. K. (1992). ALCOVE: an exemplar-based connectionist model of category learning. Psychological review, 99(1), 22.
implemented in matlab - https://github.com/noconaway/ALCOVE
Love, B. C., Medin, D. L., & Gureckis, T. M. (2004). SUSTAIN: a network model of category learning. Psychological review, 111(2), 309.
It's possible for support vector machines via a modified version of LIBSVM (by the official maintainers, cfr. here). This should probably be available in one of the many LIBSVM wrappers (e.g. kernlab, e1071). As a last resort you can always use LIBSVM's C interface through R.
