Logistic regression tutorial code. Full article available at http://datascienceplus.com/perform-logistic-regression-in-r/

logistic_regression.R

# Load the raw training data and replace missing values with NA
training.data.raw <- read.csv('train.csv',header=T,na.strings=c(""))

# Output the number of missing values for each column
sapply(training.data.raw,function(x) sum(is.na(x)))

# Quick check for how many different values for each feature
sapply(training.data.raw, function(x) length(unique(x)))

# A visual way to check for missing data
library(Amelia)
missmap(training.data.raw, main = "Missing values vs observed")

# Subsetting the data
data <- subset(training.data.raw,select=c(2,3,5,6,7,8,10,12))

# Substitute the missing values with the average value
data$Age[is.na(data$Age)] <- mean(data$Age,na.rm=T)

# R should automatically code Embarked as a factor(). A factor is R's way of dealing with
# categorical variables
is.factor(data$Sex)         # Returns TRUE
is.factor(data$Embarked)    # Returns TRUE

# Check categorical variables encoding for better understanding of the fitted model
contrasts(data$Sex)
contrasts(data$Embarked)

# Remove rows (Embarked) with NAs
data <- data[!is.na(data$Embarked),]
rownames(data) <- NULL

# Train test splitting
train <- data[1:800,]
test <- data[801:889,]

# Model fitting
model <- glm(Survived ~.,family=binomial(link='logit'),data=train)
summary(model)

# Analysis of deviance
anova(model,test="Chisq")

# McFadden R^2
library(pscl)
pR2(model)

#-------------------------------------------------------------------------------
# MEASURING THE PREDICTIVE ABILITY OF THE MODEL

# If prob > 0.5 then 1, else 0. Threshold can be set for better results
fitted.results <- predict(model,newdata=subset(test,select=c(2,3,4,5,6,7,8)),type='response')
fitted.results <- ifelse(fitted.results > 0.5,1,0)

misClasificError <- mean(fitted.results != test$Survived)
print(paste('Accuracy',1-misClasificError))

# Confusion matrix
library(caret)
confusionMatrix(data=fitted.results, reference=test$Survived)

library(ROCR)
# ROC and AUC
p <- predict(model, newdata=subset(test,select=c(2,3,4,5,6,7,8)), type="response")
pr <- prediction(p, test$Survived)
# TPR = sensitivity, FPR=specificity
prf <- performance(pr, measure = "tpr", x.measure = "fpr")
plot(prf)

auc <- performance(pr, measure = "auc")
auc <- [email protected][[1]]
auc

Thanks for the code and tutorial!

@ harigovind-s-menon
because he did not need the passenger id(the first column) in the analytics so he removed it

#Please try the below codes for contrasts:
contrasts(as.factor(data$Sex))
contrasts(as.factor(data$Embarked)

Hi,

is this code for logistic regression as statistical modell or for machine learning. I need it for statistical modell, because I did my work with Machine learning and I would to model my dataset with normale logistic regression to compare with 3 machine learning methods.
Can you help me please?
Best Regards
Arezoo

Hello Arezoo,
The code is for logistic regression model and can be used as one of the simplest machine learning tools. There is a bit of information on this link you may find helpful: https://www.r-bloggers.com/how-to-perform-a-logistic-regression-in-r/ . The good thing is with this method you can get the accuracy to compare with other methods, vs. simple linear regression. You may also want to consider these models for comparative work: Decision Tree, KNN, SVM, ANN, and Naïve Bayes. Good luck to you!
Shahryar

thanks

sir,
i got the same accuracy as you did but
when i write code of confusion matrix it shows the following error in r

confusionMatrix(data=fitted.results, reference=test$Survived)

Error in confusionMatrix(data = fitted.results, reference = test$Survived) :
unused arguments (data = fitted.results, reference = test$Survived)

kindly reply

Hi, i managed to get the error corrected by putting it in a "table": see below
confusionMatrix(table(data=fitted.results, reference=test$Survived))

sir,
i got the same accuracy as you did but
when i write code of confusion matrix it shows the following error in r

confusionMatrix(data=fitted.results, reference=test$Survived)

Error in confusionMatrix(data = fitted.results, reference = test$Survived) :
unused arguments (data = fitted.results, reference = test$Survived)

kindly reply

Hi, i managed to get the error corrected by putting it in a "table": see below
confusionMatrix(table(data=fitted.results, reference=test$Survived))

yes, it is working thanks :)

Copy mine and see if it works

…

On Wed, 14 Oct 2020, 12:29 pm sagnik99rocks, ***@***.***> wrote: ***@***.**** commented on this gist. ------------------------------ sir, i got the same accuracy as you did but when i write code of confusion matrix it shows the following error in r confusionMatrix(data=fitted.results, reference=test$Survived) Error in confusionMatrix(data = fitted.results, reference = test$Survived) : unused arguments (data = fitted.results, reference = test$Survived) kindly reply Hi, i managed to get the error corrected by putting it in a "table": see below confusionMatrix(table(data=fitted.results, reference=test$Survived)) yes, it is working thanks :) — You are receiving this because you commented. Reply to this email directly, view it on GitHub <https://gist.github.com/ac92e7c017aecff216fd#gistcomment-3488715>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AQO7QCPGFNP4EDQURTHH54DSKVVQTANCNFSM4JQSGHVQ> .

Pls help when i run the confusionMatrix, using table()) error msg i got was all arguments must have the same length. Pls help advise whats that problem

Fix for the error (data and reference should be factors with the same levels.)in

confusionMatrix(data=fitted.results, reference=test$Survived)

do:

install.packages("e1071")
confusionMatrix(data=as.factor(fitted.results), reference=as.factor(test$Survived))