mksamelson

    url = 'https://www.bloomberg.com/markets/earnings-calendar/us'

    headless_proxy = "127.0.0.1:3128"

    proxy = Proxy({
        'proxyType': ProxyType.MANUAL,
        'httpProxy': headless_proxy,
        'ftpProxy': headless_proxy,
        'sslProxy': headless_proxy,
        'noProxy': ''

mksamelson

#I use the Caret package and the xgbtree model

#I use 5 fold cross validation

clf.train = train(churn~.,
                  data=train_standardized,
                  method = "xgbTree",
                  preProcess = NULL,
                  metric = "Accuracy",
                  maximize=TRUE,

mksamelson

# Visualize Target

cbind(freq=table(churn.info$churn), percentage=prop.table(table(churn.info$churn))*100)

   freq percentage
0 52562   79.07746
1 13907   20.92254

mksamelson

#Change all negative values in the user_spendings columns to "NA"

client.info.dt[,user_spendings:=lapply(.SD,function(x)ifelse(x>=0,x,NA)),.SDcols="user_spendings"]

hist(client.info.dt[month=="8",na.omit(user_spendings)],
     labels=TRUE,
     xlim=c(0,60),
     breaks=seq(0,1300,by=10),
     main="user_spendings - August, 2013",
     xlab="user_spendings",

mksamelson

#user_spendings 

#This variable shows some negative spendings - so we remove as errors

hist(client.info.dt[month=="8",user_spendings],
     main="user_spendings - August, 2013",
     xlab="user_spendings",
     col = "green",
     border = "blue")

mksamelson

#Subset the data

client.info.jun = client.info.dt[month=="6"]
client.info.jul = client.info.dt[month=="7"]
client.info.aug = client.info.dt[month=="8"]

#Tag the variables

names(client.info.jun)[5:length(client.info.jun)] = paste(names(client.info.jun)[5:length(client.info.jun)],"6",sep="_")
names(client.info.jul)[5:length(client.info.jul)] = paste(names(client.info.jul)[5:length(client.info.jul)],"7",sep="_")

mksamelson

#Basic inspection

str(client.info)
dim(client.info)
sum(is.na(client.info))
summary(client.info)

mksamelson

#Load historical data and churn information files for basic inspection

client.info = read.csv("clientinfo.csv")
churn.info = read.csv("accountinfo.csv")

mksamelson

#Training set will be 80% of the data and Validation Set 20% of the data
#Class distribution preserved 

set.seed(0)
in.train <- createDataPartition(client.info.merged$churn, p=0.8, list=FALSE)
summary(factor(client.info.merged$churn))
ytra <- client.info.merged$churn[in.train]; summary(factor(ytra))
ytst <- client.info.merged$churn[-in.train]; summary(factor(ytst))

train = client.info.merged[in.train]

mksamelson

    party            member         state cash_avail
   (fctr)            (fctr)        (fctr)      (dbl)
1       D   Murphy, Patrick       Florida    4308768
2       D Van Hollen, Chris      Maryland    3663652
3       D  Duckworth, Tammy      Illinois    3653185
4       D    Doggett, Lloyd         Texas    3318858
5       D   Neal, Richard E Massachusetts    2730845
6       D  Sanchez, Loretta    California    2121240
7       D         Kind, Ron     Wisconsin    2107641
8       D  Kennedy, Joe III Massachusetts    2098488