Question 1 & 2: Develop the Model & Assess Predictor
Significance
Step 1: Install and load required libraries
#install.packages("readxl")
#install.packages("Hmisc")
#install.packages("pscl")
#if(!require(pROC)) install.packages("pROC")
library(readxl) #allows us to import excel files
library(Hmisc) #allows us to call the correlation function
##
## Attaching package: 'Hmisc'
## The following objects are masked from 'package:base':
##
## format.pval, units
library(pscl) #allows us to call the pseudo R-square package to evaluate our model
## Classes and Methods for R originally developed in the
## Political Science Computational Laboratory
## Department of Political Science
## Stanford University (2002-2015),
## by and under the direction of Simon Jackman.
## hurdle and zeroinfl functions by Achim Zeileis.
library(pROC) #allows us to run the area under the curve (AUC) package to get the plot and AUC score
## Type 'citation("pROC")' for a citation.
##
## Attaching package: 'pROC'
## The following objects are masked from 'package:stats':
##
## cov, smooth, var
Step 2: Import & clean the data
tire_df <- read_excel("Class Exercise 15_TireRatings.xlsx")
ti_df <- subset(tire_df, select = -c(Tire)) #drop irrelevant column
Step 3: Summarize the data
head(ti_df)
## # A tibble: 6 × 4
## Wet Noise Buy_Again Purchase
## <dbl> <dbl> <dbl> <dbl>
## 1 8 7.2 6.1 0
## 2 8 7.2 6.6 1
## 3 7.6 7.5 6.9 1
## 4 6.6 5.4 6.6 0
## 5 5.8 6.3 4 0
## 6 6.3 5.7 4.5 0
Data Description: A description of the features are presented in the table below.
Variable | Definition
------------|--------------
1. Wet | Tire Wet Traction performance rating (1: Buy again and 0: Won't buy again)
2. Noise | Tire noise level generated rating (1: Buy again and 0: Won't buy again)
3. Purchase | Average of tire purchases made
summary(ti_df)
## Wet Noise Buy_Again Purchase
## Min. :4.300 Min. :3.600 Min. :1.400 Min. :0.0000
## 1st Qu.:6.450 1st Qu.:6.000 1st Qu.:3.850 1st Qu.:0.0000
## Median :7.750 Median :7.100 Median :6.150 Median :0.0000
## Mean :7.315 Mean :6.903 Mean :5.657 Mean :0.4412
## 3rd Qu.:8.225 3rd Qu.:7.925 3rd Qu.:7.400 3rd Qu.:1.0000
## Max. :9.200 Max. :8.900 Max. :8.900 Max. :1.0000
Interpretation: The median purchase is 0, with a median of 1 meaning customers purchased tires based off of their Wet and Noise performance ratings.
Step 4: Feature selection (i.e., correlation analysis)
corr <- rcorr(as.matrix(ti_df))
corr
## Wet Noise Buy_Again Purchase
## Wet 1.00 0.76 0.91 0.74
## Noise 0.76 1.00 0.83 0.72
## Buy_Again 0.91 0.83 1.00 0.83
## Purchase 0.74 0.72 0.83 1.00
##
## n= 68
##
##
## P
## Wet Noise Buy_Again Purchase
## Wet 0 0 0
## Noise 0 0 0
## Buy_Again 0 0 0
## Purchase 0 0 0
Interpretation: All the predictors are significant with the target variable (i.e., Purchase). There's no multicollinearity in the data.
Step 5: Build the logistic regression model
model <- glm(Purchase ~ Wet + Noise, family = binomial, data = ti_df)
summary(model)
##
## Call:
## glm(formula = Purchase ~ Wet + Noise, family = binomial, data = ti_df)
##
## Coefficients:
## Estimate Std. Error z value Pr(>|z|)
## (Intercept) -39.4982 12.4779 -3.165 0.00155 **
## Wet 3.3745 1.2641 2.670 0.00760 **
## Noise 1.8163 0.8312 2.185 0.02887 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## (Dispersion parameter for binomial family taken to be 1)
##
## Null deviance: 93.325 on 67 degrees of freedom
## Residual deviance: 27.530 on 65 degrees of freedom
## AIC: 33.53
##
## Number of Fisher Scoring iterations: 8
Interpretation: ALl the independent variables were significant (p-value < 0.05)
QUestion 3: Overall Model Significance
Likelihood Ratio Test
# Fit a null model
null_model <- glm(Purchase ~ 1, data = ti_df, family = binomial)
# Perform likelihood ratio test
anova(null_model, model, test = "Chisq")
## Analysis of Deviance Table
##
## Model 1: Purchase ~ 1
## Model 2: Purchase ~ Wet + Noise
## Resid. Df Resid. Dev Df Deviance Pr(>Chi)
## 1 67 93.325
## 2 65 27.530 2 65.795 5.162e-15 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Interpretation: The inclusion of Wet and Noise as predictors in our LR model does indeed significantly predict the likelihood
of customers purchasing the Tires based off of their wet and noise performances, relative to a model that predicts purchases based solely on the mean of observed outcomes (i.e. null model).
Pseudo-R-squared
pR2(model)
## fitting null model for pseudo-r2
## llh llhNull G2 McFadden r2ML r2CU
## -13.7649516 -46.6623284 65.7947536 0.7050093 0.6199946 0.8305269
Interpretation: A McFadden R-squared of 0.71 means that our LR model explains about 71% of the variability in the outcome relative to a model with no predictors. This is not considered a good
fit since the value is too high from where a useful model usually lies (0.2 to 0.4).
Area Under the curve (AUC)
The Area Under the Curve (AUC) score represents the ability of the model to correctly classify customers that will purchase tires from Tire Rack and those who will not.
# Compute ROC Curve and the AUC score
roc_curve <- roc(ti_df$Purchase, fitted(model))
## Setting levels: control = 0, case = 1
## Setting direction: controls < cases
plot(roc_curve)
auc(roc_curve)
## Area under the curve: 0.9741
Interpretation: An AUC score of 0.97 indicates that the LR model has a high level of accuracy in predicting customer tire purchases.