machine-learning-car-data.Rmd

---
title: "Car Sales Prediction"
output: github_document
---

```{r setup, include=FALSE}
library(tidyverse)
library(Metrics)

#Loading the dataset
carDF <- read.csv("Data/car_data.csv", header = TRUE)
carDF$IsBadBuy <- as.factor(carDF$IsBadBuy)

#Confusion Matrix Calculation
confusion_matrix <- function(preds, actuals, cutoff){
  
  classifications <- ifelse(preds>cutoff,1,0)
  
  ##careful with positives and negatives here!
  confusion_matrix <- table(actuals,classifications)
}


class_performance <- function(confusion_matrix){

  TP <- confusion_matrix[2,2]
  TN <- confusion_matrix[1,1]
  FP <- confusion_matrix[1,2]
  FN <- confusion_matrix[2,1]
  
  ##accuracy = total number of correct classifications/total number of classifications
  acc <- (TP+TN)/(TP+TN+FP+FN)
  
  ##TPR = Percent of actual positives identified as such (sensitivity)
  tpr <- TP/(TP+FN)
  
  ##TNR = Percent of actual negatives identified as such (specificity)
  tnr <- TN/(TN+FP)
  
  ##I'll leave it as an exercise for you to compute the other basic confusion matrix metrics
  
  ##return the list of metrics you want
  return(c(acc, tpr, tnr))
}

```

## Part 1: Data Preparation 

1) Setting the seed and partitioning the data. The proportion of observations in the training data set should be 70%. The remaining 30% of observations should be in the test data set. 

```{r}

#Setting the seed
set.seed(71923)

#Train Test Split
train_insts = sample(nrow(carDF), .7*nrow(carDF))
carDFTrain <- carDF[train_insts,]
carDFTest <- carDF[-train_insts,]

```

## Part 2: Exploratory Data Analysis

1) Boxplots of VehOdo and VehicleAge (broken up by values of IsBadBuy).

```{r}
#Plotting the box plots
ggplot(carDFTrain, aes( x = IsBadBuy, y = VehOdo)) + geom_boxplot()
ggplot(carDFTrain, aes(x = IsBadBuy, y = VehicleAge)) + geom_boxplot()
```

We can see that bad buys have a higher vehicle age.

2) Report a two-way table for count of IsBadBuy broken up by Make (i.e. what are the numbers of each make of cars that are good vs. bad buys?)

```{r}

#Two-way table 
table(carDFTrain$Make, carDFTrain$IsBadBuy)

```

For Ford vehicles bad buys are really high than good buys. But we are not really controlling for other variables hence cannot make a concrete argument here.

## Part 3: Logistic Regression

1) Training a logistic regression model of IsBadBuy on the demographic variables using the training data set. 

```{r}
#Logistic Regression
RegressionPart4 <- glm(IsBadBuy ~ Auction + VehicleAge + Make +Color + 
                        WheelType + VehOdo + Size + MMRAcquisitionAuctionAveragePrice + 
                        MMRAcquisitionRetailAveragePrice, data = carDFTrain, family = "binomial")
summary(RegressionPart4)
ConfusionMatrixTwo <- confusion_matrix(preds = predict(RegressionPart4, carDFTest, type = "response"), actuals = carDFTest$IsBadBuy, cutoff = .5)
ConfusionMatrixTwo
```