# How can I implement logistic regression in live decision system?

I got the equation for logistic reg, and I am comfortable with the result. Let's say logit(p), ln(p/q), or the model is something like
$$\text{logit}(p) = b+a_1X_1 + a_2X_2 + a_3X_3$$ For example --> $b = 10 , a_1 = 0.5 , a_2 = 0.6 , a_3 =0.7$

So my equation is $\text{logit(p)} = 10 + 0.5X_1 + 0.6X_2 + 0.7X_3 Let's say if I want my probability to be 40% ($p = 0.4\$ ) to be my cut off rate in the system.

How can I use this in the live system?

I'm thinking of giving the developer this logit equation but use in the live system this equation and block account that have fall below success rate threshold?

• Could you please elaborate on what specifically it means to "use" a regression model in a "live system"? – whuber Jan 18 '13 at 21:54

Do the logistic regression in R, make the data into an R object, The estimated regression is then an object, and prediction can be done using it via the predict() function. Put this data object and regression model object into an R package. Then the developer programming the live system need only load this package!