Prison Guard’s Dilemma: Optimal Inmate Assignment by Multi-Objective MILO

he Pennsylvania Department of Correction operates 29 correctional facilities (prisons) and about 50,000 prisoners (inmates) each year. The assignments of inmates to appropriate correctional facilities is a complex task. Well over 80 rules need to be considered. Many of them, such as the security of prison units, yield hard constraints; while others, such assigning the inmates to prisons close to their home, are arranged in a preference hierarchy because it is impossible to satisfy all for all inmates.

We are giving an overview of the complexity of the problem; discuss the data/rule collection phase of the project by using decision trees; discuss a how the MILO model is developed by using a weighted penalty objective function. Finally we discuss further extensions of the model, including waiting lists for mental/educational/job training programs, and transfers between facilities.
Based on joint work with L. Plebani, M. Shahabsafa, G. Wilson and K. Bucklen.