A have a data frame with shops with a number of independent variables (all categorical), I'm trying to find a decent model to predict a dependent variable that is ordered and categorical (ie volume of foot traffic: "low", "medium", "high"). The dependent variable is defined for about 20% of the observations, and I want to predict for the missing 80%.
What models/techniques would you suggest?
For now, I started by learning some association rules with arules
(manual here):
require(arules)
train # my data frame with the complete observations
target_field = 'foot.traffic'
# filter rules to observe only the relevant ones
appearance_list = list(rhs=paste0(target_field,"=", unique(train[,target_field])))
rules <- arules::apriori(train,
parameter = list( support=0.1, confidence=0.2 ),
appearance = appearance_list)
rules <- sort(rules, by="lift")
inspect(rules)
Now I can inspect these rules, and they generally make sense. They mostly refer to low
and high
cases, indicating that rules for medium
are harder to identify.
How can I apply them back to the training set to see how well they predict the foot.traffic
variable? Is there a good off-the-shelf strategy to apply the rules?