Iris Classifier — Clean, Reproducible Notebook

Short: A tidy, well-documented Iris classifier notebook (Jupyter .ipynb) that demonstrates correct ML workflow: EDA → train/test split → scaling → model training → hyperparameter tuning → evaluation — with emphasis on avoiding overfitting and showing reproducible results.

---

Project overview

This repo contains a single-end project intended for learning and portfolio use. It builds notebook into a proper, reproducible pipeline using scikit-learn and documents findings clearly.

Goals

• Demonstrate correct ML workflow and reproducibility.
• Train a robust classifier on the Iris dataset.
• Evaluate with proper metrics.
• Explain limitations and next steps (no fake 100% claims).

---

Dataset

• File included: dataSets/IRIS.csv (150 rows, 4 features + target).
• Source: Classic Iris dataset (UCI / scikit-learn).
• Target classes: Iris-setosa, Iris-versicolor, Iris-virginica.

---

Repository structure

.
├── README.md
├── dataSets/
│   └── IRIS.csv
├── notebook/
│   └── iris_classifier.ipynb
└── requirements.txt
 

---

Dependencies

Create a virtual environment and install dependencies:

python -m venv .venv
source .venv/bin/activate      # macOS / Linux
.venv\Scripts\activate         # Windows

pip install -r requirements.txt

---

How to run

1. Activate the virtual environment .
2. Start Jupyter:

jupyter lab        # or jupyter notebook

3. Open notebook/iris_classifier.ipynb.
4. Run cells top-to-bottom.

---

Notebook highlights (what it contains)

• Data loading & sanity checks — head, dtypes, class balance, missing values.

• Visual EDA — pairplot / feature distributions to show class separability.

• Train/Test split — train_test_split(..., stratify=y, random_state=42, test_size=0.2).

• Scaling — StandardScaler fit on X_train then transform test set.

• Model training:

  • RandomForestClassifier with GridSearchCV hyperparameter tuning.

• Evaluation:

  • accuracy_score, confusion_matrix, classification_report.

• Feature importance (Random Forest) visualized and printed.

• Short discussion about overfitting, model choice, and limitations.

• Reproducibility: random_state=42 used where applicable.

---

Reproduce the exact reported results

The notebook includes the code that produced these results on the current run (example output included for transparency):

Accuracy: 0.966667  (96.6667%)

Classification Report:
                 precision    recall  f1-score   support

    Iris-setosa       1.00      1.00      1.00        10
Iris-versicolor       1.00      0.90      0.95        10
 Iris-virginica       0.91      1.00      0.95        10

       accuracy                           0.97        30
      macro avg       0.97      0.97      0.97        30
   weighted avg       0.97      0.97      0.97        30

NOTE: Because the dataset is small, exact test-split accuracy can vary slightly depending on the random seed or the train/test split. Use the cross-validation cell to get a more stable performance estimate.

---

Recommended commands & snippets (copy-paste)

Train with GridSearch (already in notebook):

from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import GridSearchCV

param_grid = {
    'n_estimators': [50, 100, 200],
    'max_depth': [None, 5, 10],
    'min_samples_split': [2, 5, 10]
}

grid = GridSearchCV(RandomForestClassifier(random_state=42), param_grid, cv=5)
grid.fit(X_train, y_train)

best = grid.best_estimator_
y_pred = best.predict(X_test)

Key findings & interpretation

• Iris-setosa is linearly separable → always near-perfect precision/recall.
• Iris-versicolor and Iris-virginica overlap in feature space → most errors happen between these two classes.
• Final tuned Random Forest gave ~96.7% accuracy on the held-out 20% test split — excellent for a toy dataset but do not overinterpret small improvements beyond this point: risk of overfitting is high.
• Cross-validation should be used to report robust performance (mean ± std).

---

Limitations

• Small dataset (150 samples) → high variance in single-split metrics.
• Classic toy problem: not representative of real-world messy data.
• Model interpretability: Random Forest gives feature importances but is still an ensemble (less interpretable than a single tree/logistic model).

---

Next steps (if you want to beef it up)

• Run StratifiedKFold cross-validation and report averaged metrics.
• Try SVC with RBF kernel and XGBoost / LightGBM (requires adding their packages) for comparison.
• Add a small holdout or bootstrap strategy for extra robustness.
• Create a requirements.txt lockfile with pinned versions (e.g., pip freeze > requirements.txt).
• Add a short results/ folder with PNGs: confusion matrix, feature importances, and pairplot.

---

Contact

• LinkedIn: www.linkedin.com/in/murali-krishna-m893
• Email: murali.krishna1591@gmail.com

---