There are many possible answers for this, I am strongly supportive of the conceptual one. If your input is costly, it still needs to be on the input side. But what you REALLY are asking is about the "quality problem". The true multidimensionality of output is the real challenge to good DEA or other inefficiency modeling. What you need to do is to figure out how to measure that quality or efficiency as you are calling it, or perhaps what you're calling an "input" really is a quality output, but whichever it is, you need to do the careful conceptualizing to make your problem as you estimate it correct. In most cases, this is far more difficult than it first appears, and you're asking a good initial question to start. Once the real output quality measure is in there, AND the costly input correctly, then you can measure how it works for each DMU.

Thanks for response. I guess i didnt make it clear. the problem is that there is a input which needs to be maximised for better efficiency. Other inputs are Normal i. e. they have to be maximised for higher efficiency. output is also normal and need to be maximised. How can i model this for imput oriented model.

I agree with James Burgess. However, please see:

Seiford and Zhu (2002). "Modeling undesirable factors in efficiency evaluation". European Journal of Operational Research 142, 16-20. Their method can be applied to situations when some inputs need to be increased to improve the performance.

Thanks a lot. the reference quoted was exactly what I was looking for.