BUILDING ID prediction using only WAPs

kmeans clustering magik

set.seed(13)

# load libraries
library(party)
library(randomForest)
library(rpart)
library(rpart.plot)

#load data
maxWAP_build <- read.csv(file="~/2810_uc3_R/T4C4/maxWAP_build.csv")

head(maxWAP_build)

# WAP clustering 

## one may rebuild the clusters or use the ones already saved in the maxWAP_build from previous runs. 
## not super important at this point, possibly later for sake of consistency would keep the clusters saved.

# matWAP <- as.matrix(WAPs) # create a WAP matrix to apply k-means clustering

# 3 clusters
#kmeansWAPS <- kmeans(matWAP, centers=3, iter.max = 1000, algorithm = c("Hartigan-Wong", "Lloyd", "Forgy","MacQueen"))
# K-means clustering with 3 clusters of sizes 8445, 8995, 2497
#maxWAP_build$cluster <- kmeansWAPS$cluster # save cluster vector as a variable in maxWAP_build
#head(maxWAP_build)

# repeat steps for 5 clusters. 
# kmeansWAPS5 <- kmeans(matWAP, centers=5, iter.max = 1000, algorithm = c("Hartigan-Wong", "Lloyd", "Forgy","MacQueen"))
# maxWAP_build$cluster5 <- kmeansWAPS5$cluster

# Make sure cluster vectors c(1,2,3) and c(1,2,3,4,5), as well as buildingID are saved as factors in the maxWAP_build.
maxWAP_build$df.BUILDINGID <- as.factor(maxWAP_build$df.BUILDINGID)
maxWAP_build$cluster <- as.factor(maxWAP_build$cluster)
maxWAP_build$cluster5 <- as.factor(maxWAP_build$cluster5)

# head(maxWAP_build)

controls = rpart.control(minsplit=20, maxdepth = 7) # using maxVal as predictor
fitMVt <- rpart(df.BUILDINGID~ df.maxVal,
 data = maxWAP_build,
 method="class",
 control= controls)
rpart.plot(fitMVt)

# Create the forest.
fitMVrf <- randomForest(df.BUILDINGID ~ df.maxVal, 
           data = maxWAP_build)

# View the forest results.
print(fitMVrf) 

Call:
 randomForest(formula = df.BUILDINGID ~ df.maxVal, data = maxWAP_build) 
               Type of random forest: classification
                     Number of trees: 500
No. of variables tried at each split: 1

        OOB estimate of  error rate: 50.83%
Confusion matrix:
    1   2    3 class.error
1 376 111 4762   0.9283673
2 153 948 4095   0.8175520
3 294 718 8480   0.1066161

controls = rpart.control(minsplit=20, maxdepth = 7) # using cluster 3 as predictor
fit3t <- rpart(df.BUILDINGID~ cluster,
 data = maxWAP_build,
 method="class",
 control= controls)
rpart.plot(fit3t)

# Create the forest.
fit3rf <- randomForest(df.BUILDINGID ~ cluster, 
           data = maxWAP_build)

# View the forest results.
print(fit3rf) 

Call:
 randomForest(formula = df.BUILDINGID ~ cluster, data = maxWAP_build) 
               Type of random forest: classification
                     Number of trees: 500
No. of variables tried at each split: 1

        OOB estimate of  error rate: 16.3%
Confusion matrix:
     1    2    3 class.error
1 2497 2752    0  0.52429034
2    0 5196    0  0.00000000
3    0  497 8995  0.05235988

controls = rpart.control(minsplit=20, maxdepth = 7) # using cluster 3 and cluster 5 as predictors
fit35t <- rpart(df.BUILDINGID~ cluster+cluster5,
 data = maxWAP_build,
 method="class",
 control= controls)
rpart.plot(fit35t)

# Create the forest.
fit35rf <- randomForest(df.BUILDINGID ~ cluster+cluster5, 
           data = maxWAP_build)

# View the forest results.
print(fit35rf) 

Call:
 randomForest(formula = df.BUILDINGID ~ cluster + cluster5, data = maxWAP_build) 
               Type of random forest: classification
                     Number of trees: 500
No. of variables tried at each split: 1

        OOB estimate of  error rate: 3.92%
Confusion matrix:
     1    2    3 class.error
1 4957  292    0  0.05562964
2    0 5196    0  0.00000000
3    0  490 9002  0.05162242

controls = rpart.control(minsplit=20, maxdepth = 7) # using cluster 3, cluster 5 and maxVal as predictors
fitt <- rpart(df.BUILDINGID~ .,
 data = maxWAP_build,
 method="class",
 control= controls)
rpart.plot(fitt)

# Create the forest.
fitrf <- randomForest(df.BUILDINGID ~ ., 
           data = maxWAP_build)

# View the forest results.
print(fitrf) 

Call:
 randomForest(formula = df.BUILDINGID ~ ., data = maxWAP_build) 
               Type of random forest: classification
                     Number of trees: 500
No. of variables tried at each split: 1

        OOB estimate of  error rate: 3.92%
Confusion matrix:
     1    2    3 class.error
1 4957  292    0  0.05562964
2    0 5196    0  0.00000000
3    0  490 9002  0.05162242