R Lab: Classifying Penguins
Nicole Hayek
2022-12-04
SETUP
library(tidyverse)## ── Attaching packages ─────────────────────────────────────── tidyverse 1.3.2 ──
## ✔ ggplot2 3.3.6 ✔ purrr 0.3.4
## ✔ tibble 3.1.8 ✔ dplyr 1.0.10
## ✔ tidyr 1.2.1 ✔ stringr 1.4.1
## ✔ readr 2.1.2 ✔ forcats 0.5.2
## ── Conflicts ────────────────────────────────────────── tidyverse_conflicts() ──
## ✖ dplyr::filter() masks stats::filter()
## ✖ dplyr::lag() masks stats::lag()library(caTools)
library(caret)## Loading required package: lattice
##
## Attaching package: 'caret'
##
## The following object is masked from 'package:purrr':
##
## liftlibrary(MASS)##
## Attaching package: 'MASS'
##
## The following object is masked from 'package:dplyr':
##
## selectlibrary(nnet)
library(rpart)
library(rpart.plot)library(palmerpenguins)
penguins %>%
drop_na() -> clean.data
set.seed(284) # pick your own seed number (I used my abc123 number.)
sample_split <- sample.split(Y = clean.data$species, SplitRatio = 0.5)
train_set <- subset(x = clean.data, sample_split == TRUE)
test_set <- subset(x = clean.data, sample_split == FALSE)Analysis
Exercise 1:Fit and assess a logistic regression model using the multinom(), predict(), and confusionMatrix() functions. How accurate is this model?
summary(
logreg.model <- multinom(factor(species) ~ bill_length_mm + bill_depth_mm +
flipper_length_mm + body_mass_g,
data=train_set, maxit=50)
)## # weights: 18 (10 variable)
## initial value 183.468252
## iter 10 value 12.367957
## iter 20 value 2.244638
## iter 30 value 0.545504
## iter 40 value 0.008076
## iter 50 value 0.000437
## final value 0.000437
## stopped after 50 iterations## Call:
## multinom(formula = factor(species) ~ bill_length_mm + bill_depth_mm +
## flipper_length_mm + body_mass_g, data = train_set, maxit = 50)
##
## Coefficients:
## (Intercept) bill_length_mm bill_depth_mm flipper_length_mm
## Chinstrap -3.424724 54.200630 -7.862102 -9.445153
## Gentoo -2.062269 8.234405 -23.872150 -1.546746
## body_mass_g
## Chinstrap -0.11831060
## Gentoo 0.08214113
##
## Std. Errors:
## (Intercept) bill_length_mm bill_depth_mm flipper_length_mm
## Chinstrap 0.20111311 1.92589567 40.19717111 9.9985935
## Gentoo 0.00139627 0.02550289 0.03699088 0.1024792
## body_mass_g
## Chinstrap 0.5459752
## Gentoo 0.1025563
##
## Residual Deviance: 0.0008749982
## AIC: 20.00087predicted_class <- predict(logreg.model, test_set)
confusionMatrix(test_set$species, predicted_class)## Confusion Matrix and Statistics
##
## Reference
## Prediction Adelie Chinstrap Gentoo
## Adelie 71 1 1
## Chinstrap 4 30 0
## Gentoo 0 0 59
##
## Overall Statistics
##
## Accuracy : 0.9639
## 95% CI : (0.923, 0.9866)
## No Information Rate : 0.4518
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.943
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: Adelie Class: Chinstrap Class: Gentoo
## Sensitivity 0.9467 0.9677 0.9833
## Specificity 0.9780 0.9704 1.0000
## Pos Pred Value 0.9726 0.8824 1.0000
## Neg Pred Value 0.9570 0.9924 0.9907
## Prevalence 0.4518 0.1867 0.3614
## Detection Rate 0.4277 0.1807 0.3554
## Detection Prevalence 0.4398 0.2048 0.3554
## Balanced Accuracy 0.9623 0.9691 0.9917There are 344 observations with 4 numeric predictor variables and 2-factor predictor variables (sex and island). The variable year is ignored in this model. This model has a 96.39% accuracy which proves it is very accurate.
Exercise 2: Fit and assess a linear discriminant analysis model using the lda(), predict(), and confusionMatrix() functions. How accurate is this model?
CART.model <- rpart(species ~ bill_length_mm + bill_depth_mm + flipper_length_mm + body_mass_g,
data = train_set, method = "class")
rpart.plot(CART.model)
As proven above the accuracy of the model is still quite high.
Exercise 3: Which variables were (automatically) selected for this model? If you look at the models above, you can see that the variables bill_depth_mm and body_mass_g did not make the cut. However, the automatically selected variables are bill length and flipper length.
Exercise 4: How accurate is this model?
preds <- predict(CART.model, newdata = test_set, type = "class")
confusionMatrix(test_set$species, preds)## Confusion Matrix and Statistics
##
## Reference
## Prediction Adelie Chinstrap Gentoo
## Adelie 72 1 0
## Chinstrap 5 26 3
## Gentoo 0 0 59
##
## Overall Statistics
##
## Accuracy : 0.9458
## 95% CI : (0.8996, 0.9749)
## No Information Rate : 0.4639
## P-Value [Acc > NIR] : < 2.2e-16
##
## Kappa : 0.9139
##
## Mcnemar's Test P-Value : NA
##
## Statistics by Class:
##
## Class: Adelie Class: Chinstrap Class: Gentoo
## Sensitivity 0.9351 0.9630 0.9516
## Specificity 0.9888 0.9424 1.0000
## Pos Pred Value 0.9863 0.7647 1.0000
## Neg Pred Value 0.9462 0.9924 0.9720
## Prevalence 0.4639 0.1627 0.3735
## Detection Rate 0.4337 0.1566 0.3554
## Detection Prevalence 0.4398 0.2048 0.3554
## Balanced Accuracy 0.9619 0.9527 0.9758For a sample such as this, this model also indicates a high level of accuracy! the level of accuracy that is indicated is = 0.9458, which is very high as well and is in accordance with the other models.
Exercise 5: How do the models compare? (Which is more accurate?). Can you suggest a reason why they differ?
Although the accuracy is high for all of the above, the highest would be with the Logistic Regression with a 0.9639. A suggestive reason of why that may be the case is because the linear discriminant analysis requires more assumptions about the underlying data.The logistic regression is the more flexible approach in the event that these assumptions are violated.