GATech.r

#Read in data
dataAdmit<-as.data.frame(read.csv("fall16.csv.txt"))

#Clean data for ones with decisions
dataAdmitClean<- dataAdmit[!(is.na(dataAdmit$Decision) | dataAdmit$Decision==""), ]

#Rather than try and replace the NA's...for no other reason than time, 
#I'll just remove them. In practice I might try regressing to predict what
#their values should be. I'd feel more comfortable with more features. 
dataAdmitClean<-dataAdmitClean[complete.cases(dataAdmitClean),]

#Now we have only complete data on which to train a model.
#We have so few data points that are complete so the validity will be questionable.
#As the semesters progress and as reports come in, and as feature reporting becomes standard,
#I expect accuracy will increase. I take issue with "Selective school" being <25% admission,
#when some top CS schools have a higher overall admission but do not release their CS admission rate.
#Ranking of school in CS nationally might be a better feature. Might include it later.


#***NOTE***: The subsequent model is pretty much useless.
#I started playing around with an SVM and a decision tree briefly before I realized that there was insufficient
#data about rejections specifically to make a meaningful predictive model.
#Perhaps with more data about rejections, meaningful predictions could be made. 

install.packages('e1071', dependencies = TRUE)
library("e1071")
dataAdmitClean
#Here I assumed dates were not significant factors, everybody would have 3 letters in, and that selective/BigTech was sufficient
admit_model <- svm(formula=Decision ~ GPA+ Selective+ BigTech, data=dataAdmitClean)
summary(admit_model)