Do you have attribute-value pairs in your training/test set?

If so, the following steps are used to remove the errors due to the

missing attributes or the missing values of the training data set.

Step 1(Missing attributes): In this case, the missing attribute is added and its value is obtained based on step 2.

Step 2 (Missing values): the abundant value in the related class is selected for symbolic missing value (e.g. true/false) and the average of the values in the related class is calculated and set to the missing numeric values.

Hey, here is my answer. I don't know all the details in your problem. but let's assume you have a number of N patients for which you have measured the number of WBC among other parameters. I would simple remove the patients with missing WBC from the analysis. If you don't remove them and this number is 20-30% of your total number then your end results will be affected by the procedure to correct for missing values.

What you can do  is to make a feature selection on the data that you have (there are free matlab  tools on the internet). Then you can see what is the rank of WBC, if the rank is low you can simple dismiss them from the analysis.

good luck

Look at Daphne Kollers' book : "Probabilistic Graphical Models", chapter 19 : "Partially Observed Data" and : https://class.coursera.org/pgm/lecture/100 "Learning with Incomplete Data". The reason and the way missing data were created influences the decision to "exclude" them, or continue your computations considering all possible values the missing variables could take.

There are multiple ways of handling missing data. The most common and easiest but not the best is ommitting missing observations. This is not recommended.

One very important point here is the reason of being a missing value. If there is a special reason behind being missing, which has a link to the outcome or exposure of the study you should be more carfull. Because then simply replacing missing values or not including them in the analysis will introduce bias into your result.

I am not familiar with matlab unfortunately. But missing data imputation techniques may have been intdoruced in matlab.

Read more about impuation here: 

I have found teh following link for Matlab which might be helpful. Check it your self.

http://www.cc.gatech.edu/~hpark/softwareMVE.html

Regards,

Mehdi

Dear Negar thank you so much for your answer. I didn't understand your question. What do you mean by Attribute-Value pairs?

As you know when we are talking about a data set, we mean that we have a number of samples (here patients) each of which has a number of attributes (such as WBC, RBC, Blood pressure,... in this study). When I'm talking about missing values in a particular attribute, let's say in WBC, I mean that there is a number of patients for whom there is no information in the attribute WBC.

Based on this definition I can not understand what did you mean by partitioning missing attributes and missing values. 

Missing Values means incomplete dataset....and followd by this Incomplete Information......Like the same way Incomple Knowledge... You can apply set theoritic approachs like Rough Sets and Other Rough Set varients like Variable Precesion Rough Set , Neighbourhood Rough Set etc...These Set theoric approaches have the ability to remove incompleteness of a dataset. So, I believe if you go thrugh the Rough Set related Research Papers you will find solution of your problem.

I sent you a pdf file that it can help you

Dear mahnaz

you are completely right

examples for attributes: age, hypertension, gender, etc

and values are the information about attributes

according to your explanation of your project, you are using attribute-value system too

so you can use those 2 methods in order to deal with that problem

good luck

In this case usually bayesian network function is acceptable

bayesian network in such problem use maximum expectation method that you can easly find more about it by search on the internet

finally if you know algorithm you don't have any problem in implementation whether in MATLAB or other software and programming language

Thank you Reza Sadeghi for your answer and wise suggestion. 

Of course I know how to program and what is sufficient for programming! I only mentioned that point to tell my friends that I do not decide to use statistical tools such as SPSS which can easily handle missing data by their built-in functions and algorithms. But I will be glad if I hear about any MATLAB tools relevant to this problem.

Regards  

Mahnaz -

As Mehdi noted, "One very important point here is the reason of being a missing value." There is "ignorable nonresponse," and "nonignorable nonresponse," the ideas of which would apply to any missing data, i suppose, as the first just means that there is no special reason for the nonresponse, so it was as if a random selection is missing, and the later means those missing data are - in a way - 'unrepresented' by the data collected. I am only familiar with continuous data, but there are several methods to try to account for the missing data by imputation, and then there is (re)weighting of the data if you are trying to look at an estimate for a finite population. To impute a mean would usually mean you are expecting that the data were missing at random, so that would be true for ignorable nonresponse, but you could group your data by -as Mehdi said - the reason for the missing data. Often these reasons will actually give you different response rates for different groups of data that you may identify, called "response propensity groups." If you group your data that way, and impute or reweight within groups, then you could obtain more 'representative' results. As I stated, I only have expertise with continuous data, but there are a number of possible ways to impute data. (I think a previous answer above noted 'nearest neighbor,' for one.) But whenever possible, I think regression (so-called"prediction") is very useful. It means that you can use data you already have, and the relationships between these data, to good advantage. It is also extremely useful that you can estimate the "variance of the prediction error" (see econometrics textbooks, such as those by GS Maddala) to have an idea of some of the uncertainty involved. It will tell you much about when you have a good imputed data, and when you do not. Of course sometimes you can find much more highly correlated regressor (x) data for your y value of interest than in other cases. It is very advantageous to be able to estimate uncertainty. (In SAS PROC REG, the square root of the estimated variance of the prediction error is "STDI.") Note that the individual standard errors for the prediction errors will somewhat 'cancel,' to a degree, so that the variance of the prediction error for an estimated total may be much better than one might expect. (Both individual and group variance of the prediction error are discussed in the early pages of the attached.)

I highly recommend scatterplots for comparing/studying your continuous data. 

So, for your continuous data, at least, some of the terminology above might, if you search the Internet, give you some ideas as to what might be the best way(s) for you to handle your missing data.

Cheers - Jim   

Article Using Prediction-Oriented Software for Survey Estimation

One can use rough set techniques to add missing attribute values. These are called incomplete information systems.The paper by Kryzik(spelling not correct) in 1998 can be of help and also other papers in that direction can be seen.

you can use filter from the Weka data mining tool. it is free.As far as method is concerned it depends your requirement.you still fill the missing value by:

1. global constant

2. nearest neighbor

3. binning method.

for detail plz refer to Han and kamber book. data preprocessing.