Online Community Feature.Rmd

---
output:
  pdf_document: default
  html_document: default
---
Load Libraries

```{r}
rm(list=ls())
library(readxl)
library(ggplot2)
library(Hmisc)
library(MASS)
## library(caret)
library(regclass)
library(ISLR)
library(boot)
library(vcd)
library(pROC)
library (ROCR)
```


After loading the libraries, load the excel file containing data on Defaults

```{r}

##set your working directory

#setwd("/path/to/my/directory")
library("readxl")
##setwd("")
mydata = read_excel("Data2.xlsx")

summary(mydata) #summary of the data
```
You could also use

```{r}
str(mydata)
```


What do you notice? Looks like the Variables default and student are *character* and not *numeric*. Let's look at the values they take.

```{r}
head(mydata)
```

head alone might not display _all possible values_. Let's find all the possible values

```{r}
levels(as.factor(mydata$`Joined?`))
levels(as.factor(mydata$`Churned at 3 months after launch of the online community`))

```


Looks like both have only two levels and are Yes/No variables. Let's convert all No to 0 and Yes to 1. To do so, we create two new variables in the same mydata table. The two variables are labelled default2 and student2.


```{r}
mydata$churn<- ifelse(mydata$`Churned at 3 months after launch of the online community` == "No",0,1)
mydata$join <- ifelse(mydata$`Joined?` == "No",0,1)
```

Confirm the change by looking at the data again. And get the summary.

```{r}
head(mydata)

levels(as.factor(mydata$churn))
levels(as.factor(mydata$join)) #if you wanted to recheck the levels
```
```{r}
summary(mydata)
```

Now let's code the model. We want to predict default. The independent variables in our dataset are 'student=Yes/No', 'balance' and 'income'. The default variable is discrete and has only two levels: {0,1}. Unlike in a regression, where the dependent variable is continuous, default being discrete does not lend itself to standard regression analysis.

```{r}
#Logistic Regression Model Estimation
mylogit<-glm(`Churned at 3 months after launch of the online community`~ `Average Spend Last 3 months of Life with the firm`+ `Joined?`+ `Customer Age with Firm at time of launching the online community` ,data=mydata,family=binomial(link="logit"))
```


Let's output the results

```{r}
#coefficients
summary(mylogit)
```

While the p-values indicate confidence, we could also output the 2.5% and 97.5% confindence intervals
```{r}
#Confidence Intervals
Confidence=confint(mylogit) 
Confidence
```

Another way to interpret the parameter estimates would be to compute the Odds Ratios

```{r}
#Odds Ratio Calculation, including confidence intervals
oddsr=round(exp(cbind(OddsRatio=coef(mylogit),confint(mylogit))),4)

oddsr

##CLV Calculation 
## we calculate L -> the xepected lifetime of a consumer 
churn_yes_joined <- data.frame(subset(mydata, mydata$`Churned at 3 months after launch of the online community`== 1 &  mydata$`Joined?`== 1 ))


churn_total_joined <- data.frame(subset(mydata,  mydata$`Joined?`== 1 ))



nr <-nrow(churn_yes_joined)
dr <- nrow(churn_total_joined)
churn_of_those_who_joined <- nr/ dr

churn_of_those_who_joined

r <- 1- churn_of_those_who_joined ## retention rate of those who joined
r
L <- 1/ (1-r) ## avg expected customer life time 
m <- mean(churn_total_joined$Average.Spend.Last.3.months.of.Life.with.the.firm*3)*0.50
CLV <- m*L ## CLV of customers who joined 
CLV 

churn_yes_not_joined <- data.frame(subset(mydata, mydata$`Churned at 3 months after launch of the online community`== 1 &  mydata$`Joined?`== 0 ))


churn_total_not_joined <- data.frame(subset(mydata,  mydata$`Joined?`== 0 ))

nr_nj <-nrow(churn_yes_not_joined)
dr_nj <- nrow(churn_total_not_joined)
churn_of_those_who_didnt_join <- nr_nj/ dr_nj

churn_of_those_who_didnt_join
r_nj <- 1- churn_of_those_who_didnt_join ## retention rate of those who joined
L_nj <- 1/ (1-r_nj) ## avg expected customer life time 
m_nj <- mean(churn_total_not_joined$Average.Spend.Last.3.months.of.Life.with.the.firm*3) *0.50
CLV_nj <- m_nj*L_nj ## CLV of customers who joined 

CLV_nj

r_nj

###### CLV forevery customer 
churn_yes <- data.frame(subset(mydata, mydata$`Churned at 3 months after launch of the online community`== 1 ))
churn <- nrow(churn_yes)/ nrow(mydata)

R <- 1 - churn
L <- 1/(1-R)


mydata$CLV <- ((mydata$`Average Spend Last 3 months of Life with the firm`*0.50) * L)

CLV_Joined <- lm(mydata$CLV ~ mydata$`Joined?`)
summary (CLV_Joined)

```


```{r}
# The average revenue and CLV increased with decreased retention, where the increase in CLV can be attributed to the growth in revenue. However, in the long run as the examination period increases, the CLV being very sensitive to retention rate could be negatively affected by the low retention value. In order to further validate this claim, we recommend the management to run an A/B test of the app on a small sample size and check if the online community causes the change in retention.
```