Polytechnic University of Valencia Congress, CARMA 2016 - 1st International Conference on Advanced Research Methods and Analytics

Font Size: 
Weighting machine learning solutions by economic and institutional context for decision making
Jose A. Alvarez-Jareño, Jose M. Pavía

Last modified: 23-06-2017


It  is  quite  common  that  machine  learning  approaches  reach  high  accuracy forecast  rates  in  imbalanced  datasets.  However,  the  results  in  the  category with  few  instances  are  usually  low.  This  paper  seeks  to  improve  the  results obtained applying different techniques (such as bagging, boosting or random forests)  with  the  inclusion  of  cost  matrices.  We  propose  applying  the  actual costs  incurred  by  the  company  for  misclassification  of  instances  as  a  cost matrix.  This  approach,  along  with  an  economic  analysis  of  the  different solutions,  makes  it  possible  to  incorporate  a  business  perspective  in  the decision  making  process.  The  approach  is  tested  on  a  publicly  available dataset. In our example, the best ratings are obtained by combining the cost matrix  with  random  forests.  However,  our  analysis  shows  that  the  best technical  solution  is  not  always  the  best  economical  solution  available.  A company  cannot  always  implement  the  optimal  solution,  but  has  to  adopt  a solution  constrained  by  its  social,  institutional  and  economic  context.  Once an  economic  analysis  is  carried  out,  it  seems  the  final  decision  of  the company will depend on its economic situation and its institutional policy.

Full Text: PDF