R-Script

########################
## Load Packages & Data ## 
########################
library(readr)
library(caret)
library(C50)
library(doParallel)
library(tidyr)
library(RMariaDB)
library(RMySQL)
library(ggplot2)

## Import data
RAWdata <- read.csv("trainingData.csv")

########################
## parallel processing ## 
########################

# Find how many cores are on your machine
detectCores() # Result = 4
# Create Cluster with desired number of cores. Don't use them all! Your computer is running other processes. 
cl <- makeCluster(2)
# Register Cluster
registerDoParallel(cl)
# Confirm how many cores are now "assigned" to R and RStudio
getDoParWorkers() # Result 2 
# Stop Cluster. After peorming your tasks, stop your cluster. 
#stopCluster(cl)

####################
## Pre-processing ## 
####################

# Verify data type
str(RAWdata) #19937 observations
# Verify data type first 9 rows
str(RAWdata[,1:9])
summary(RAWdata[,1:9]) 

# Missing values? 
is.na(RAWdata) #false

# Basic histogram buildings
ggplot(RAWdata, aes(x=BUILDINGID)) + geom_histogram()

# Identify columns with no variance using nearZeroVar()  
nzvMetrics <- nearZeroVar(RAWdata, saveMetrics = TRUE)
nzvMetrics

# remove columns with no variance using nearZeroVar()  
RAWdata = subset(RAWdata, select = -c(WAP003,WAP004,WAP092,WAP093,WAP094,WAP095,WAP152,WAP158,WAP159,WAP160,WAP215,WAP217,WAP226,WAP227,WAP238,WAP239,WAP240,WAP241) )


#########################
## Feature Engineering ## 
#########################

# Group dataset by building number
FEdata.BLD1 <- RAWdata[RAWdata$BUILDINGID == 0, ]
FEdata.BLD2 <- RAWdata[RAWdata$BUILDINGID == 1, ]
FEdata.BLD3 <- RAWdata[RAWdata$BUILDINGID == 2, ]

## Loc_UID2 = LONGITUDE, LATITUDE, FLOOR producing 933 levels

# Combine LONGITUDE, LATITUDE, Floor 
FEdata2 <- unite(RAWdata, "Loc_UID2", remove = TRUE, sep = ".", c("LONGITUDE","LATITUDE", "FLOOR"))
Location2_Bld1 <- unite(FEdata.BLD1, "Loc_UID2", remove = TRUE, sep = ".", c("LONGITUDE","LATITUDE", "FLOOR"))
Location2_Bld2 <- unite(FEdata.BLD2, "Loc_UID2", remove = TRUE, sep = ".", c("LONGITUDE","LATITUDE", "FLOOR"))
Location2_Bld3 <- unite(FEdata.BLD3, "Loc_UID2", remove = TRUE, sep = ".", c("LONGITUDE","LATITUDE", "FLOOR"))

# Convert Loc_UID2 to factor 
FEdata2$Loc_UID2 <- as.factor(FEdata2$Loc_UID2)
Location2_Bld1$Loc_UID2 <- as.factor(Location2_Bld1$Loc_UID2)
Location2_Bld2$Loc_UID2 <- as.factor(Location2_Bld2$Loc_UID2)
Location2_Bld3$Loc_UID2 <- as.factor(Location2_Bld3$Loc_UID2)

# verify data type
str(FEdata2[,1:9])

# Remove extra columns
FEdata2 = subset(FEdata2, select = -c(BUILDINGID,SPACEID,RELATIVEPOSITION,USERID,PHONEID,TIMESTAMP) )
Location2_Bld1 = subset(Location2_Bld1, select = -c(BUILDINGID,SPACEID,RELATIVEPOSITION,USERID,PHONEID,TIMESTAMP) )
Location2_Bld2 = subset(Location2_Bld2, select = -c(BUILDINGID,SPACEID,RELATIVEPOSITION,USERID,PHONEID,TIMESTAMP) )
Location2_Bld3 = subset(Location2_Bld3, select = -c(BUILDINGID,SPACEID,RELATIVEPOSITION,USERID,PHONEID,TIMESTAMP) )

# -- Remove attribute with 0 variance from Building 1 -- #
# Identify columns with no variance using nearZeroVar()  
nzvMetrics <- nearZeroVar(Location2_Bld1, saveMetrics = TRUE)
nzvMetrics

# remove columns with no variance using nearZeroVar()  
Location2_Bld1 = subset(Location2_Bld1, select = -c(WAP200, WAP203, WAP204, WAP206, WAP208, WAP209, WAP213, WAP214, WAP216, WAP218, WAP219, WAP220, WAP222, WAP223, WAP228, WAP229, WAP230, WAP231, WAP232, WAP233, WAP234, WAP235, WAP236, WAP237, WAP242, WAP243, WAP244, WAP245, WAP246, WAP247, WAP249, WAP250, WAP254, WAP255, WAP256, WAP257, WAP258, WAP259, WAP260, WAP261, WAP262, WAP263, WAP274, WAP277, WAP278, WAP279, WAP282, WAP284, WAP286, WAP288, WAP293, WAP295, WAP296, WAP301, WAP303, WAP304, WAP307, WAP310, WAP311, WAP312, WAP313, WAP314, WAP315, WAP316, WAP317, WAP318, WAP329, WAP332, WAP333, WAP334, WAP335, WAP338, WAP340, WAP342) )

# -- Remove attribute with 0 variance from Building 2 -- #
# Identify columns with no variance using nearZeroVar()  
nzvMetrics <- nearZeroVar(Location2_Bld2, saveMetrics = TRUE)
nzvMetrics

# remove columns with no variance using nearZeroVar()  
Location2_Bld2 = subset(Location2_Bld2, select = -c(WAP001,WAP002,WAP005,WAP007,WAP009,WAP013,WAP014,WAP018,WAP019,WAP020,WAP025,WAP026,WAP031,WAP032,WAP033,WAP034,WAP039,WAP040,WAP041,WAP042,WAP047,WAP048,WAP053,WAP054,WAP057,WAP058,WAP060,WAP061,WAP062,WAP063,WAP064,WAP065,WAP066,WAP067,WAP068,WAP070,WAP071,WAP072,WAP073,WAP074,WAP075,WAP076,WAP079,WAP080,WAP081,WAP087,WAP096,WAP097,WAP098,WAP099,WAP100,WAP117,WAP118,WAP128,WAP138,WAP139) )

# -- Remove attribute with 0 variance from Building 3 -- #
# Identify columns with no variance using nearZeroVar()  
nzvMetrics <- nearZeroVar(Location2_Bld3, saveMetrics = TRUE)
nzvMetrics

# remove columns with no variance using nearZeroVar()  
Location2_Bld3 = subset(Location2_Bld3, select = -c(WAP001,WAP002,WAP007,WAP008,WAP009,WAP010,WAP013,WAP014,WAP015,WAP016,WAP017,WAP019,WAP020,WAP021,WAP022,WAP023,WAP024,WAP025,WAP026,WAP027,WAP028,WAP029,WAP030,WAP031,WAP032,WAP033,WAP034,WAP035,WAP036,WAP037,WAP038,WAP039,WAP040,WAP041,WAP042,WAP043,WAP044,WAP045,WAP046,WAP047,WAP048,WAP049,WAP050,WAP051,WAP052,WAP053,WAP054,WAP055,WAP056,WAP057,WAP058,WAP071,WAP072,WAP075,WAP076,WAP079,WAP080,WAP081,WAP086,WAP088,WAP089,WAP090,WAP091,WAP100,WAP101,WAP102,WAP103,WAP104,WAP105,WAP106,WAP119,WAP120,WAP123,WAP124,WAP125,WAP126,WAP129,WAP130,WAP133,WAP134,WAP135,WAP136,WAP137) )

# review variable count
str(Location2_Bld1) #429 variables
str(Location2_Bld2) #447 variables
str(Location2_Bld3) #420 variables

#################
## Modeling RF ## 
#################

# --- Model: All 3 building RF -- #

# Configure Train test split 
set.seed(123)
trainSize<-round(nrow(FEdata2)*0.8) 
testSize<-nrow(FEdata2)-trainSize
trainSize # 13956
testSize # 5981
training_indices<-sample(seq_len(nrow(FEdata2)),size =trainSize)
trainSet<-FEdata2[training_indices,]
testSet<-FEdata2[-training_indices,] 

# 5 fold cross validation
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1)

# configure train algo
system.time(rfFitBuild <- train(Loc_UID2~., data = trainSet, method = "rf", trControl=fitControl, tuneLength = 1))
# Run time 2.7 hours

#training results
rfFitBuild
# Accuracy     Kappa 


# ---  Predict all 3 Buildings RF --- # 

rfPred <- predict(rfFitBuild, newdata = testSet)
#To see the initial row of the predicited data
head(rfPred)
summary(rfFitBuild)

#To determine how the model prioritized each feature
varImp(rfFitBuild) # most important -- 

rfClasses_i <- predict(rfFitBuild, newdata = testSet)
str(rfClasses_i)

postResample(rfPred, testSet$Loc_UID2)
# Accuracy     Kappa 
# 0.7965223   0.7962457

summary(rfPred)
plot(rfPred)

# --- Model: Building 1 RF -- #

set.seed(123)
trainSize<-round(nrow(Location2_Bld1)*0.8) 
testSize<-nrow(Location2_Bld1)-trainSize
trainSize # 4199
testSize # 1050
training_indices<-sample(seq_len(nrow(Location2_Bld1)),size =trainSize)
trainSet<-Location2_Bld1[training_indices,]
testSet<-Location2_Bld1[-training_indices,] 
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1)
system.time(rfFitBld1 <- train(Loc_UID2~., data = trainSet, method = "rf", trControl=fitControl, tuneLength = 1))
# Run time: 17 minutes
rfFitBld1
# Accuracy     Kappa 
# 0.7677928  0.7667815

# ---  Predict Building 1 RF --- # 

rfPredBLD1 <- predict(rfFitBld1, newdata = testSet)
head(rfPredBLD1)
summary(rfFitBld1)
varImp(rfFitBld1) # most important -- [WAP051]
rfClassesBLD1 <- predict(rfFitBld1, newdata = testSet)
str(rfClassesBLD1)
postResample(rfPredBLD1, testSet$Loc_UID2)
# Accuracy     Kappa 
# 0.7704762 0.7694471 
summary(rfPredBLD1)
plot(rfPredBLD1)


# --- Model: Building 2 RF -- #

set.seed(123)
trainSize<-round(nrow(Location2_Bld2)*0.8) 
testSize<-nrow(Location2_Bld2)-trainSize
trainSize # 4157
testSize # 1039
training_indices<-sample(seq_len(nrow(Location2_Bld2)),size =trainSize)
trainSet<-Location2_Bld2[training_indices,]
testSet<-Location2_Bld2[-training_indices,] 
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1)
system.time(rfFitBld2 <- train(Loc_UID2~., data = trainSet, method = "rf", trControl=fitControl, tuneLength = 1))
# Run time: 22  minutes
rfFitBld2
# Accuracy     Kappa 
#0.8487728  0.847977

# ---  Predict Building 2 RF --- # 

rfPredBLD2 <- predict(rfFitBld2, newdata = testSet)
head(rfPredBLD2)
summary(rfFitBld2)
varImp(rfFitBld2) # most important -- 
rfClassesBLD2 <- predict(rfFitBld2, newdata = testSet)
str(rfClassesBLD2)
postResample(rfPredBLD2, testSet$Loc_UID2)
# Accuracy     Kappa 
#0.8487728  0.847977
summary(rfPredBLD2)
plot(rfPredBLD2)


# --- Model: Building 3 RF -- #

set.seed(123)
trainSize<-round(nrow(Location2_Bld3)*0.8) 
testSize<-nrow(Location2_Bld3)-trainSize
trainSize # 7594
testSize # 1898
training_indices<-sample(seq_len(nrow(Location2_Bld3)),size =trainSize)
trainSet<-Location2_Bld3[training_indices,]
testSet<-Location2_Bld3[-training_indices,] 
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1)
system.time(rfFitBld3 <- train(Loc_UID2~., data = trainSet, method = "rf", trControl=fitControl, tuneLength = 1))
# Run time: 45 minutes
rfFitBld3
# Accuracy     Kappa 
# 0.8142142  0.8136415

# ---  Predict: Building 3 RF --- # 

rfPredBLD3 <- predict(rfFitBld3, newdata = testSet)
head(rfPredBLD3)
summary(rfFitBld3)
varImp(rfFitBld3) # most important -- 
rfClassesBLD3 <- predict(rfFitBld3, newdata = testSet)
str(rfClassesBLD3)
postResample(rfPredBLD3, testSet$Loc_UID2)
# Accuracy     Kappa 
# 0.8113804 0.8107842 
summary(rfPredBLD3)
plot(rfPredBLD3)

#-----------------------------------------------------------------
#################
## Modeling C5 ## 
#################

# --- Model: All 3 Buildings C5 -- #

set.seed(123)
trainSize<-round(nrow(FEdata2)*0.8) 
testSize<-nrow(FEdata2)-trainSize
trainSize # 13956
testSize # 5981
training_indices<-sample(seq_len(nrow(FEdata2)),size =trainSize)
trainSet<-FEdata2[training_indices,]
testSet<-FEdata2[-training_indices,] 
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1)
system.time(c5FitBuild <- train(Loc_UID2~., data = trainSet, method = "C5.0", trControl=fitControl, tuneLength = 1))
# Run time  15 minutes
c5FitBuild
# Accuracy     Kappa 
#0.5845119  0.5839530

# ---  Predict: All 3 Buildings C5 --- # 

c5Pred <- predict(c5FitBuild, newdata = testSet)
head(c5Pred)
summary(c5FitBuild)
varImp(c5FitBuild) # most important -- 
c5Classes_i <- predict(c5FitBuild, newdata = testSet)
str(c5Classes_i)
postResample(c5Pred, testSet$Loc_UID2)
# Accuracy     Kappa 
#0.5895335 0.5889930 
summary(c5Pred)
plot(c5Pred)

# --- Model: Building 1 C5 -- #

set.seed(123)
trainSize<-round(nrow(Location2_Bld1)*0.8) 
testSize<-nrow(Location2_Bld1)-trainSize
trainSize # 4199
testSize # 1050
training_indices<-sample(seq_len(nrow(Location2_Bld1)),size =trainSize)
trainSet<-Location2_Bld1[training_indices,]
testSet<-Location2_Bld1[-training_indices,] 
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1)
system.time(c5FitBld1 <- train(Loc_UID2~., data = trainSet, method = "C5.0", trControl=fitControl, tuneLength = 1))
# Run time: 2 minutes
c5FitBld1
# Accuracy     Kappa 
#0.5780753  0.5762808

# ---  Predict: Building 1 C5 --- # 

c5PredBLD1 <- predict(c5FitBld1, newdata = testSet)
head(c5PredBLD1)
summary(c5FitBld1)
varImp(c5FitBld1) # most important -- 
c5ClassesBLD1 <- predict(c5FitBld1, newdata = testSet)
str(c5ClassesBLD1)
postResample(c5PredBLD1, testSet$Loc_UID2)
# Accuracy     Kappa 
#0.5733333 0.5715055 
summary(c5PredBLD1)
plot(c5PredBLD1)

# --- Model: Building 2 C5 -- #

set.seed(123)
trainSize<-round(nrow(Location2_Bld2)*0.8) 
testSize<-nrow(Location2_Bld2)-trainSize
trainSize # 4157
testSize # 1039
training_indices<-sample(seq_len(nrow(Location2_Bld2)),size =trainSize)
trainSet<-Location2_Bld2[training_indices,]
testSet<-Location2_Bld2[-training_indices,] 
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1)
system.time(c5FitBld2 <- train(Loc_UID2~., data = trainSet, method = "C5.0", trControl=fitControl, tuneLength = 1))
# Run time:  2 minutes
c5FitBld2
# Accuracy     Kappa 
#0.6892071  0.6875879

# ---  Predict: Building 2 C5 --- # 

c5PredBLD2 <- predict(c5FitBld2, newdata = testSet)
head(c5PredBLD2)
summary(c5FitBld2)
varImp(c5FitBld2) # most important -- 
c5ClassesBLD2 <- predict(c5FitBld2, newdata = testSet)
str(c5ClassesBLD2)
postResample(c5PredBLD2, testSet$Loc_UID2)
# Accuracy     Kappa 
#0.7054860 0.7038425 
summary(c5PredBLD2)
plot(c5PredBLD2)


# --- Model: Building 3 C5 -- #

set.seed(123)
trainSize<-round(nrow(Location2_Bld3)*0.8) 
testSize<-nrow(Location2_Bld3)-trainSize
trainSize # 7594
testSize # 1898
training_indices<-sample(seq_len(nrow(Location2_Bld3)),size =trainSize)
trainSet<-Location2_Bld3[training_indices,]
testSet<-Location2_Bld3[-training_indices,] 
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1)
system.time(c5FitBld3 <- train(Loc_UID2~., data = trainSet, method = "C5.0", trControl=fitControl, tuneLength = 1))
# Run time: 5 minutes
c5FitBld3
# Accuracy     Kappa 
#0.5753771  0.5740953

# ---  Predict: Building 3 C5 --- # 

c5PredBLD3 <- predict(c5FitBld3, newdata = testSet)
head(c5PredBLD3)
summary(c5FitBld3)
varImp(c5FitBld3) # most important -- 
c5ClassesBLD3 <- predict(c5FitBld3, newdata = testSet)
str(c5ClassesBLD3)
postResample(c5PredBLD3, testSet$Loc_UID2)
# Accuracy     Kappa 
#0.5648051 0.5634856 
summary(c5PredBLD3)
plot(c5PredBLD3)


#-----------------------------------------------------------------

##################
## Modeling KNN ## 
##################

# --- Model: All 3 Buildings KNN -- #

set.seed(123)
trainSize<-round(nrow(FEdata2)*0.8) 
testSize<-nrow(FEdata2)-trainSize
trainSize # 13956
testSize # 5981
training_indices<-sample(seq_len(nrow(FEdata2)),size =trainSize)
trainSet<-FEdata2[training_indices,]
testSet<-FEdata2[-training_indices,] 
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1)
system.time(knnFitBuild <- train(Loc_UID2~., data = trainSet, method = "knn"))
# Run time 48 minutes
knnFitBuild
# Accuracy     Kappa 
# 0.4928898  0.4922288

# ---  Predict: All 3 Buildings KNN -- #

knnPred <- predict(knnFitBuild, newdata = testSet)
head(knnPred)
summary(knnFitBuild)
varImp(knnFitBuild) # most important -- 
Classes_i <- predict(knnFitBuild, newdata = testSet)
str(knnClasses_i)
postResample(knnPred, testSet$Loc_UID2)
# Accuracy     Kappa 
#0.5917071 0.5911681 
summary(knnPred)
plot(knnPred)

# --- Model: Building 1 KNN -- #

set.seed(123)
trainSize<-round(nrow(Location2_Bld1)*0.8) 
testSize<-nrow(Location2_Bld1)-trainSize
trainSize # 4199
testSize # 1050
training_indices<-sample(seq_len(nrow(Location2_Bld1)),size =trainSize)
trainSet<-Location2_Bld1[training_indices,]
testSet<-Location2_Bld1[-training_indices,] 
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1)
system.time(knnFitBld1 <- train(Loc_UID2~., data = trainSet, method = "knn"))
# Run time: 6  minutes
knnFitBld1
# Accuracy     Kappa 
#0.4777012  0.4755206

# ---  Predict: Building 1 KNN -- #

knnPredBLD1 <- predict(knnFitBld1, newdata = testSet)
head(knnPredBLD1)
summary(knnFitBld1)
varImp(knnFitBld1) # most important -- 
knnClassesBLD1 <- predict(knnFitBld1, newdata = testSet)
str(knnClassesBLD1)
postResample(knnPredBLD1, testSet$Loc_UID2)
# Accuracy     Kappa 
#0.5590476 0.5571618 
summary(knnPredBLD1)
plot(knnPredBLD1)

# --- Model: Building 2 KNN -- #

set.seed(123)
trainSize<-round(nrow(Location2_Bld2)*0.8) 
testSize<-nrow(Location2_Bld2)-trainSize
trainSize # 4157
testSize # 1039
training_indices<-sample(seq_len(nrow(Location2_Bld2)),size =trainSize)
trainSet<-Location2_Bld2[training_indices,]
testSet<-Location2_Bld2[-training_indices,] 
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1)
system.time(knnFitBld2 <- train(Loc_UID2~., data = trainSet, method = "knn"))
# Run time: 5 minutes
knnFitBld2
# Accuracy     Kappa 
#0.5574969  0.5552950

# ---  Predict: Building 2 KNN -- #

knnPredBLD2 <- predict(knnFitBld2, newdata = testSet)
head(knnPredBLD2)
summary(knnFitBld2)
varImp(knnFitBld2) # most important -- 
knnClassesBLD2 <- predict(knnFitBld2, newdata = testSet)
str(knnClassesBLD2)
postResample(knnPredBLD2, testSet$Loc_UID2)
# Accuracy     Kappa 
#0.6679500 0.6661445 
summary(knnPredBLD2)
plot(knnPredBLD2)


# --- Model: Building 3 KNN -- #

set.seed(123)
trainSize<-round(nrow(Location2_Bld3)*0.8) 
testSize<-nrow(Location2_Bld3)-trainSize
trainSize # 7594
testSize # 1898
training_indices<-sample(seq_len(nrow(Location2_Bld3)),size =trainSize)
trainSet<-Location2_Bld3[training_indices,]
testSet<-Location2_Bld3[-training_indices,] 
fitControl <- trainControl(method = "repeatedcv", number = 10, repeats = 1) 
system.time(knnFitBld3 <- train(Loc_UID2~., data = trainSet, method = "knn"))
# Run time: 17 minutes
knnFitBld3
# Accuracy     Kappa 
#0.5239142  0.5224957           

# ---  Predict: Building 3 KNN -- #

knnPredBLD3 <- predict(knnFitBld3, newdata = testSet)
head(knnPredBLD3)
summary(knnFitBld3)
varImp(knnFitBld3) # most important -- 
knnClassesBLD3 <- predict(knnFitBld3, newdata = testSet)
str(knnClassesBLD3)
postResample(knnPredBLD3, testSet$Loc_UID2)
# Accuracy     Kappa 
#0.6138040 0.6126061 
summary(knnPredBLD3)
plot(knnPredBLD3)