Classification with naiveBayes (e1071) does not work ($levels returns NULL)
classification, machine-learning, r
Solution
Make sure that you treat class variable as factor; i.e.
nb.classifier <- naiveBayes(as.factor(class) ~ ., data = arrhythmia.training)
By the way, you don't need to exclude class variable from predict call.
Problem
I use naiveBayes (e1071 http://en.wikibooks.org/wiki/Data_Mining_Algorithms_In_R/Classification/Na%C3%AFve_Bayes) for classifying my data set (Classification class: "class" 0/1). Here is what I do: ``` library(e1071) arrhythmia <- read.csv(file="/home/.../arrhythmia.csv", head=TRUE, sep=",") #devide into training and test data 70:30 trainingIndex <- createDataPartition(arrhythmia$class, p=.7, list=F) arrhythmia.training <- arrhythmia[trainingIndex,] arrhythmia.testing <- arrhythmia[-trainingIndex,] nb.classifier <- naiveBayes(class ~ ., data = arrhythmia.training) predict(nb.classifier,arrhythmia.testing[,-260]) ``` The classifier does not work, here is what I get: ``` > predict(nb.classifier,arrhythmia.testing[,-260]) factor(0) Levels: > str(arrhythmia.training) 'data.frame': 293 obs. of 260 variables: $ age : int 75 55 13 40 44 50 62 54 30 46 ... $ sex : int 0 0 0 1 0 1 0 1 0 1 ... $ height : int 190 175 169 160 168 167 170 172 170 158 ... $ weight : int 80 94 51 52 56 67 72 58 73 58 ... $ QRSduration : int 91 100 100 77 84 89 102 78 91 70 ... $ PRinterval : int 193 202 167 129 118 130 135 155 180 120 ... # and so on (260 attributes) > str(arrhythmia.training[260]) 'data.frame': 293 obs. of 1 variable: $ class: int 1 0 1 0 0 1 1 1 1 0 ... > nb.classifier$levels NULL ``` I tried to use the included the data set (iris) and everything works fine. What's wrong with my approach?