Package importing
pacman::p_load(dplyr, broom, caTools, ggplot2, gridExtra, caret)
Dataset importing
dataset = read.delim('Restaurant_Reviews.tsv',
quote = '',
stringsAsFactors = F)
dataset.ori = read.delim('Restaurant_Reviews.tsv',
quote = '',
stringsAsFactors = F)
library('tm')
library('SnowballC')
corpus = VCorpus(VectorSource(dataset$Review))
corpus = tm_map(corpus, content_transformer(tolower))
corpus = tm_map(corpus, removeNumbers)
corpus = tm_map(corpus, removePunctuation)
corpus = tm_map(corpus, removeWords, stopwords())
corpus = tm_map(corpus, stemDocument)
corpus = tm_map(corpus, stripWhitespace)
dtm = DocumentTermMatrix(corpus)
dtm = removeSparseTerms(dtm, sparse = 0.999)
dataset = as.data.frame(as.matrix(dtm))
dataset = dataset[, -c(80, 406)]
dataset$Liked = dataset.ori$Liked
dataset$Liked = factor(dataset$Liked, levels = c(0, 1))
Dataset partitioning
set.seed(123)
split = sample.split(dataset$Liked, SplitRatio = 0.8)
training.set = subset(dataset, split == T)
test.set = subset(dataset, split == F)
Model fitting
# Logistic regression
mod.lr = glm(Liked ~ .,
data = training.set,
family = binomial)
# K-nearest neighbors
library('class')
library('scales')
accuracy.set = data.frame(k = seq(1, 20, 1), accuracy = rep(0, 20))
for (i in 1:20) {
y.pred = knn(train = training.set[, -ncol(training.set)],
test = test.set[, -ncol(test.set)],
cl = training.set[, ncol(training.set)],
k = i)
accuracy.set[i, 2] = confusionMatrix(y.pred,
test.set[, ncol(test.set)])$overall[1]
}
ggplot(data = accuracy.set, aes(x = k, y = accuracy)) +
geom_point() +
geom_line(linetype = 'dashed') +
scale_x_continuous(breaks = pretty_breaks(nrow(accuracy.set))) +
scale_y_continuous(breaks = pretty_breaks()) +
labs(title = 'Accuracy vs K-value',
subtitle = 'Best KNN',
x = 'K-value',
y = 'Accuracy')
# Support vector machine
library('e1071')
mod.svm = svm(Liked ~ .,
data = training.set,
type = 'C-classification',
kernel = 'linear')
# Kernel support vector machine
library('e1071')
mod.ksvm = svm(Liked ~ .,
data = training.set,
type = 'C-classification',
kernel = 'radial')
# Naive bayes
library('e1071')
mod.nb = naiveBayes(x = training.set[, -ncol(training.set)],
y = training.set[, ncol(training.set)])
# Decision tree - CART
library('rpart')
mod.dt.cart = rpart(Liked ~ .,
data = training.set,
method = 'class',
control = rpart.control(xval = 10))
plotcp(mod.dt.cart)
# mod.dt.cart.p = prune(mod.dt.cart, cp = 0.013)
# plotcp(mod.dt.cart.p)
# Decision tree - C5.0
library('C50')
mod.dt.c50 = C5.0(Liked ~ .,
data = training.set)
# Random forest
library('randomForest')
mod.rf = randomForest(Liked ~ .,
data = training.set,
ntree = 500)
Predication & Tunning
# Logistic regression
y.pred.prob = predict(mod.lr, type = 'response',
newdata = test.set[, -ncol(test.set)])
y.pred = factor(ifelse(y.pred.prob > 0.5, 1, 0), levels = c(0, 1))
cm = confusionMatrix(y.pred, test.set[, ncol(test.set)])
ac = format(round(cm$overall[1]*100, 2), nsmall = 2) %>% as.numeric()
f1 = format(round(cm$byClass['F1']*100, 2), nsmall = 2) %>% as.numeric
Accuracy = c(ac)
F1 = c(f1)
# eval = data.frame(Accuracy = ac, F1 = f1)
# K-nearest neighbors (k = 1)
y.pred = knn(train = training.set[, -ncol(training.set)],
test = test.set[, -ncol(test.set)],
cl = training.set[, ncol(training.set)],
k = 1)
cm = confusionMatrix(y.pred, test.set[, ncol(test.set)])
ac = format(round(cm$overall[1]*100, 2), nsmall = 2) %>% as.numeric()
f1 = format(round(cm$byClass['F1']*100, 2), nsmall = 2) %>% as.numeric
Accuracy = append(Accuracy, ac)
F1 = append(F1, f1)
# Support vector machine
y.pred = predict(mod.svm, newdata = test.set[, -ncol(test.set)])
cm = confusionMatrix(y.pred, test.set[, ncol(test.set)])
ac = format(round(cm$overall[1]*100, 2), nsmall = 2) %>% as.numeric()
f1 = format(round(cm$byClass['F1']*100, 2), nsmall = 2) %>% as.numeric
Accuracy = append(Accuracy, ac)
F1 = append(F1, f1)
# Kernel support vector machine
y.pred = predict(mod.ksvm, newdata = test.set[, -ncol(test.set)])
cm = confusionMatrix(y.pred, test.set[, ncol(test.set)])
ac = format(round(cm$overall[1]*100, 2), nsmall = 2) %>% as.numeric()
f1 = format(round(cm$byClass['F1']*100, 2), nsmall = 2) %>% as.numeric
Accuracy = append(Accuracy, ac)
F1 = append(F1, f1)
# Naive bayes
y.pred = predict(mod.nb, newdata = test.set[, -ncol(test.set)])
cm = confusionMatrix(y.pred, test.set[, ncol(test.set)])
ac = format(round(cm$overall[1]*100, 2), nsmall = 2) %>% as.numeric()
f1 = format(round(cm$byClass['F1']*100, 2), nsmall = 2) %>% as.numeric
Accuracy = append(Accuracy, ac)
F1 = append(F1, f1)
# Decision tree - CART
y.pred = predict(mod.dt.cart, newdata = test.set[, -ncol(test.set)], type = 'class')
cm = confusionMatrix(y.pred, test.set[, ncol(test.set)])
ac = format(round(cm$overall[1]*100, 2), nsmall = 2) %>% as.numeric()
f1 = format(round(cm$byClass['F1']*100, 2), nsmall = 2) %>% as.numeric
Accuracy = append(Accuracy, ac)
F1 = append(F1, f1)
# Decision tree - C5.0
y.pred = predict(mod.dt.c50, newdata = test.set[, -ncol(test.set)], type = 'class')
cm = confusionMatrix(y.pred, test.set[, ncol(test.set)])
ac = format(round(cm$overall[1]*100, 2), nsmall = 2) %>% as.numeric()
f1 = format(round(cm$byClass['F1']*100, 2), nsmall = 2) %>% as.numeric
Accuracy = append(Accuracy, ac)
F1 = append(F1, f1)
# Random forest
y.pred = predict(mod.rf, newdata = test.set[, -ncol(test.set)], type = 'class')
cm = confusionMatrix(y.pred, test.set[, ncol(test.set)])
ac = format(round(cm$overall[1]*100, 2), nsmall = 2) %>% as.numeric()
f1 = format(round(cm$byClass['F1']*100, 2), nsmall = 2) %>% as.numeric
Accuracy = append(Accuracy, ac)
F1 = append(F1, f1)
Model evaluating
eval = cbind(Accuracy, F1)
name = c('LR', 'KNN', 'SVM', 'KSVM', 'NB', 'DT-CART', 'DT-C50', 'RF')
rownames(eval) = name
eval = as.data.frame(eval)
eval[order(-eval$Accuracy, -eval$F1), , drop = F]
## Accuracy F1
## SVM 79.0 78.79
## RF 75.5 75.38
## DT-CART 71.0 74.56
## KNN 69.0 70.48
## DT-C50 64.0 71.65
## LR 54.0 51.58
## KSVM 53.0 68.03
## NB 50.5 9.17
rank = c('SVM', 'RF', 'DT-CART', 'KNN', 'DT-50', 'LR', 'KSVM', 'NB')
eval %>%
arrange(desc(Accuracy)) %>%
mutate(ML.name = factor(rank,
levels = c(rank))) %>%
ggplot() +
geom_col(aes(x = ML.name, y = Accuracy)) +
coord_cartesian(ylim = c(40, 90)) +
labs(title = 'Classification ML vs Accuracy',
subtitle = 'SVM_BEST',
x = '')
