################################################################# 
# THIS FILE CONTAINS THE CODE NECESSARY TO OBTAIN THE RESULTS   #
# OF EXPERIMENTS USED IN THE PAPER:                             #
# Torgo,L and Lopes,E.: Utility-based Fraud Detection. In       #
#   Proceedings of the 22nd IJCAI'2011                          #
#################################################################
# Author : Luis Torgo (ltorgo@dcc.fc.up.pt)      Date: Mar 2011 #
# License: GPL (>= 2)                                           #
#################################################################




library(DMwR)

# Load in the code for the UOR algorithm
source('UOR.R')

source('auxFunc.R')


# =====================================================
# Function to omitting the cost, benefit and utility
# =====================================================

omitFunc <- function(dat,percOmit=0.5) {
  omitCases <- sample(1:nrow(dat), as.integer(nrow(dat)*percOmit))
  dat[omitCases,3:5]<-NA
  dat
}


# =====================================================
# Main code
# =====================================================
 
load('artifData.Rdata')

percSup <- 0.8  # percentage of cases that are history (i.e. we know C,B,U)

set.seed(1234)
supCases <- sample(1:nrow(data),as.integer(nrow(data)*percSup))

descrVars <- 1:2

percOmit <- 0.0 # percentage of cases that are NA (C,B,U)


# Historical data set
histData <- as.data.frame(na.omit(omitFunc(data[supCases,],percOmit)))

# The set of candidate cases for inspection
inspData <- as.data.frame(data[-supCases,descrVars])


# =====================================================
# Experiments with Regression Trees(rpartXse) + ORh + linear utility

rm <- learner('rpartXse',pars=list(se=0.5))
om <- learner('orhCall',pars=list())

urank <- UOR(histData,inspData,regrLearner=rm,outLearner=om) # or the line below
save(urank, file='artDS_model_rpart_ORh.Rdata')


# =====================================================
# Experiments with NNet + LOF + linear utility

require(nnet)

rm <- learner('nnet',pars=list(linout=T,trace=F,size=10,maxit=1000))
om <- learner('lofCall',pars=list(k=30))

urank <- UOR(histData,inspData,regrLearner=rm,outLearner=om) # or the line below
save(urank,file='artDS_model_nnet_LOF.Rdata')