Iris Dataset Classification

This project models iris species based on petal length & width as well as sepal (green part underneath bloom) length & width. I classified this data based on 2 classic algorithms: K-NN & Decision Trees and evaluated their performance.

The data is sourced from the Fisher Iris dataset, 1936. It is one of the earliest known datasets used for evaluating classification methods.

Total Iris Samples: 150

• Setosa Count = 50

• Versicolor Count = 50

• Virginica Count: 50

Input data (features):

1. Sepal Length (cm)

2. Sepal Width (cm)

3. Petal Length (cm)

4. Petal Width (cm)

1. Classification Results:

Average Metrics (Stratified 5-Fold Cross-Validation):
A. 3-NN Average Metrics:

• Accuracy: 0.9533 ± 0.0340

• Precision: 0.9560 ± 0.0322

• Recall: 0.9533 ± 0.0340

• F1: 0.9532 ± 0.0341

B. Decision Tree Average Metrics:

• Accuracy: 0.9333 ± 0.0422

• Precision: 0.9432 ± 0.0324

• Recall: 0.9333 ± 0.0422

• F1: 0.9321 ± 0.0439

2. Broad data insights:

• Mean sepal length = 5.843 ± 0.828 cm

• Mean Sepal Width = 3.054 ± 0.434 cm

• Mean Petal Length = 3.759 ± 1.76 cm

• Mean Petal Width = 1.199 ± 0.763 cm

3. Species-specific Inisghts:

A. Average Petal Area

• Setosa ≈ 0.357 cm^2

• Versicolor ≈ 5.65 cm^2

• Virginica ≈ 11.2 cm^2

B. Average Sepal Area:

• Setosa = 17.1 cm^2

• Versicolor = 16.4 cm^2

• Virginica = 19.6 cm^2

1. Small test set size
   - 70-30 train-test split only allows for 45 total predictions

   -Highly subject to sampling bias
   - I counteracted somewhat using 5-fold cross-validation

2. Limited Experimentation with model types jk

   - only tests 2 models

   - maybe test Naive Bayes, LogRes, or Linear SVM in the future?