Problem with Logistic regression / classification problem

classification
logistic

#1

Dear all,

I’m scientific officer working in international organization my working nature is data analysis using SAS. Presently, I’m working on NFHS(National Family Health surveys) data. I am facing a problem in running a model on nutrition data. I want to study whether a child being weight is affected by mother and other variables.

My dependent variable is underweight (it has 4 categories) and independent variables are age, sex, weight of child at birth, mother’s education and so on. Most of the independent variables are categorical variables. So, here i’m totally confused which technique should I use?

Please give me your suggestions. looking forward for your valuables comments.

Regards,
Anil
Scientific officer
Markets, Institutes and Policies


#2

You may try those traditional methods like logistic regression (ordinal for ordered and multi-nominal for unordered dependent variable), decision tree and neural network. For some of them, you may need to transform your categorical (binary/nominal) variables using dummy indicators. For sure if you have access to SAS Enterprise Miner, it could save you a lot of time. It also allows you to compare as well as ensemble different models easily.


#3

Hi Anil,

In case all the input and output variables are categorical, you can try something as simple as an ANNOVA model. ANNOVA model will tell you the extent to which your each categorical variable influences the dependent variable.

Tavish


#4

HI tavish and Sunnylam,
Thank you for your valueable suggestions i will definitely try Anova. or can i use nonparametric test ?

regards,
Anil


#6

Hello,

You can try a “multinomial logistic regression”.
See here an example:

Thanks,
Carlos.