Data can be dirty and usually is!

Data types

1. Raw data
2. Technically correct data
3. Data that is proper for the model
4. Summarized data
5. Data with fixed format

Preprocessing Map

Data Cleaning

• Do you see any problems?

sim.dat <- read.csv("https://raw.githubusercontent.com/happyrabbit/DataScientistR/master/Data/SegData.csv ")
summary(sim.dat)

Data Cleaning

• How to deal with that?

# set problematic values as missings
sim.dat$age[which(sim.dat$age>100)]<-NA
sim.dat$store_exp[which(sim.dat$store_exp<0)]<-NA
# see the results
summary(subset(sim.dat,select=c("age","income")))

Missing Values

• Is there any auxiliary information?
• Is missing a random occurrence?
• What is the purpose of modeling?

Missing Values: median/mode

• impute() function in imputeMissings package

# save the result as another object
demo_imp<-impute(sim.dat,method="median/mode")
# check the first 5 columns, there is no missing values in other columns
summary(demo_imp[,1:5])

• preProcess() function in caret package

imp<-preProcess(sim.dat,method="medianImpute")
demo_imp2<-predict(imp,sim.dat)
summary(demo_imp2[,1:5])

Missing Values: K-nearest neighbors

• preProcess() function in caret package

imp<-preProcess(sim.dat,method="knnImpute",k=5)
# need to use predict() to get KNN result
demo_imp<-predict(imp,sim.dat)

Solve the problem

• Reason: sim.dat has non-numeric variables
• Solution:

# find factor columns
imp<-preProcess(sim.dat,method="knnImpute",k=5)
idx<-which(lapply(sim.dat,class)=="factor")
demo_imp<-predict(imp,sim.dat[,-idx])
summary(demo_imp[,1:3])

Missing Values: Bagging Tree

• Bagging (Bootstrap aggregating)
• Powerful
• Computation is much more intense than KNN

imp<-preProcess(sim.dat,method="bagImpute")
demo_imp<-predict(imp,sim.dat)
summary(demo_imp[,1:5])

Centering and Scaling

• Easy to DIY

income<-sim.dat$income
# calculate the mean of income
mux<-mean(income,na.rm=T)
# calculate the standard deviation of income
sdx<-sd(income,na.rm=T)
# centering
tr1<-income-mux
# scaling
tr2<-tr1/sdx

• Use preProcess()

sdat<-subset(sim.dat,select=c("age","income"))
# set the "method" option
trans<-preProcess(sdat,method=c("center","scale"))
# use predict() function to get the final result
transformed<-predict(trans,sdat)

Resolve Skewness

• describe(sim.dat)
• Box-Cox transformation

# select the two columns and save them as dat_bc
dat_bc<-subset(sim.dat,select=c("store_trans","online_trans"))
(trans<-preProcess(dat_bc,method=c("BoxCox")))

Use predict() to get the transformed result:

transformed<-predict(trans,dat_bc)

Resolve Outliers

• Defining outliers is hard
• Basic visualizations: box-plot, histogram and scatterplot
• Statistical methods to define outliers

Z-score and modified Z-score

• Z-score

Zi=Yi−ˉYs where ˉY and s are mean and standard deviation for Y

• Modified Z-score

Mi=0.6745(Yi−ˉY)MAD

where MAD is the median of a series of |Yi−ˉY|, called the median of the absolute dispersion

• Iglewicz and Hoaglin suggest that the points with the Z-score greater than 3.5 are possible outliers

Collinearity

• Define: corrplot()
• Algorithm to remove a minimum number of predictors to ensure all pairwise correlations are below a certain threshold

1. Calculate the correlation matrix of the predictors.
2. Determine the two predictors associated with the largest absolute pairwise correlation (call them predictors A and B).
3. Determine the average correlation between A and the other variables. Do the same for predictor B.
4. If A has a larger average correlation, remove it; otherwise, remove predictor B.
5. Repeat Step 2-4 until no absolute correlations are above the threshold.

• How to choose the threshold?

Sparse Variables

• Detecting rules:

1. The fraction of unique values over the sample size
2. The ratio of the frequency of the most prevalent value to the frequency of the second most prevalent value.

Re-encode Dummy Variables

• class.ind() from nnet package

dumVar<-class.ind(sim.dat$gender)
head(dumVar)

• dummyVars() from caret

dumMod<-dummyVars(~gender+house+income,
                  data=sim.dat,
                  # use "origional variable name + level" as new name
                  levelsOnly=F)
head(predict(dumMod,sim.dat))