For the sake of comparing results of classification with other works of literature.
Most methods that address missing data assume data is missing at random, NOT "missing" to remove "outliers".
One possible method to create a data set with missing values would be to start with a complete data set. Transcribe the data, drawing a Bernoulli random variable (rv) with probability p for each attribute of each observation. If rv==1 replace attribute value with a MISSING value/code. You could vary the Bernoulli parameter p and observe impact on your analysis / inference.
for a very systematic simulation-based way, but not easy to do: you may use R's simFrame package for this task, see
http://www.jstatsoft.org/v37/i03/
there you also find some code examples where missings are inserted. The paper should discuss the insertion of missings.
If you have adequate numbers of complete data you can use machine learning approach to predict missing fields !
Article Machine Learning Based Missing Value Imputation Method for C...
- Badges
- Science topic