Skip to content
All projects
ML· Feb 2026

Iris Flower Classification

A complete, reproducible ML pipeline that predicts an iris species from four measurements.

View code (opens in a new tab)
test accuracy
93.3%
Built with
Pythonscikit-learnpandasNumPy
Confusion matrix on the test set, 93.3% accuracy. Setosa 20 of 20 correct; versicolor 19 correct and 1 predicted as virginica; virginica 17 correct and 3 predicted as versicolor.
All 4 errors are between versicolor and virginica, whose measurements overlap.

Problem

Predict an iris flower's species — setosa, versicolor or virginica — from sepal and petal length and width, using a pipeline that is honest about its results and gives the same answer every run.

Approach

  • Stratified 60/40 train/test split so every species is equally represented in both sets.
  • StandardScaler fitted on the training data only, then applied to the test set — no data leakage.
  • Logistic regression classifier from scikit-learn.
  • Full evaluation: accuracy, confusion matrix, and per-class precision / recall / F1.
  • Fixed random seed for reproducibility.

Results

Metric Test set (60 flowers)
Accuracy 93.3%
Setosa 100% precision and recall
Versicolor F1 = 0.90
Virginica F1 = 0.89

Setosa is perfectly separable. All 4 errors are between versicolor and virginica, whose measurements overlap.

What I learned

  • The standard ML workflow end to end: load, split, scale, train, predict, evaluate.
  • Avoiding data leakage by fitting preprocessing on the training set only.
  • Why stratification matters on small datasets.
  • Accuracy isn't the whole story — the confusion matrix shows which classes get confused.

Dataset

Iris dataset — R. A. Fisher (1936), via the UCI Machine Learning Repository.