CODEWITHSUNDEEP
rapriori
Travis Heeter
Travis Heeter

Reputation: 14054

R- Is there a way to limit apriori rules by lift?

I'm looking at this data set: https://archive.ics.uci.edu/ml/machine-learning-databases/credit-screening/crx.data

I preprocessed the data:

ca.1<-read.csv("CreditApproval.csv",T,",")

# From http://stackoverflow.com/q/4787332/
remove_outliers <- function(x, na.rm = TRUE, ...) {
  qnt <- quantile(x, probs=c(.25, .75), na.rm = na.rm, ...)
  H <- 1.5 * IQR(x, na.rm = na.rm)
  y <- x
  y[x < (qnt[1] - H)] <- NA
  y[x > (qnt[2] + H)] <- NA
  y
}

ca.1$A2<-remove_outliers(ca$A2)
ca.1$A3<-remove_outliers(ca$A3)
ca.1$A8<-remove_outliers(ca$A8)
ca.1$A11<-remove_outliers(ca$A11)
ca.1$A14<-remove_outliers(ca$A14)
ca.1$A15<-remove_outliers(ca$A15)
ca.1$A2<-discretize(ca.1$A2,"frequency",categories = 6)
ca.1$A3<-discretize(ca.1$A3,"frequency",categories = 6)
ca.1$A8<-discretize(ca.1$A8,"frequency",categories = 6)
ca.1$A11<-discretize(ca.1$A11,"frequency",categories = 6)
ca.1$A14<-discretize(ca.1$A14,"frequency",categories = 6)
ca.1$A15<-discretize(ca.1$A15,"frequency",categories = 6)

ca.1<-na.omit(ca.1)

After fine tuning the support, confidence, min/maxlen I'm still getting 65 rules:

> rules<-apriori(ca.1, parameter= list(supp=0.15, conf=0.89, minlen=3, maxlen=4), appearance=list(rhs=c("class=-", "class=+"), default="lhs"))
> rules.sorted <- sort(rules, by="lift")
> inspect(rules.sorted)
     lhs                     rhs       support   confidence lift    
[1]  {A5=g,A9=t,A10=t}    => {class=+} 0.1521739 0.8974359  2.770607
[2]  {A4=u,A9=t,A10=t}    => {class=+} 0.1521739 0.8974359  2.770607
[3]  {A1=a,A9=f}          => {class=-} 0.1717391 0.9753086  1.442579
[4]  {A1=a,A9=f,A13=g}    => {class=-} 0.1608696 0.9736842  1.440176
...[65]

As you can see + rules have a greater lift, but less support and confidence than the - rules. I've been looking through the docs, and can't find any parameter to limit by lift. Is this possible? If not, what do you do in situations like this?

Upvotes: 2

Views: 2677

Answers (5)

juanbretti
juanbretti

Reputation: 438

Another way is to use arules::quality(). For example:

association.rules <- apriori(tr, parameter = list(support=0.005, confidence=0.25, minlen=3, maxlen=10))

subRules<-association.rules[quality(association.rules)$lift > 1]

This function can filter by support, confidence, coverage, lift, count.

Upvotes: 2

emmanuel agarry
emmanuel agarry

Reputation: 355

You can't limit apriori rules by lift alone. You have to get a limit by support and confidence first which you did here:

 rules<-apriori(ca.1, parameter= list(supp=0.15, conf=0.89, minlen=3, maxlen=4)

Then after that, do something like this

rulesLift <- sort(subset(rules, subset = lift < 2), by="lift") 
inspect(rulesLift)

Upvotes: 2

Denys
Denys

Reputation: 66

In arules package a special function to subset this object type is defined. In order to filter out rules with lift value less than 2 you can try the following:

subset(rules, subset = lift > 2)

Upvotes: 5

user131476
user131476

Reputation: 422

I think apriori function does not take lift as one of the parameter. I get this error if I try to set lift

Error: Invalid parameter: lift

Instead I could sort the rules by lift and pick the rules based on the lift value as follows

sort (rules, by="lift", decreasing=TRUE)

This is not a straightforward solution but a decent workaround

Upvotes: 0

Nick Knauer
Nick Knauer

Reputation: 4243

What if you tried: 

apriori(df, parameter = list(lift = 0.3, minlen =2))

You can set your minimum lift to anything in this case, just chose 0.3.

Upvotes: -1

Related Questions