Bug Prediction using KNN and Logistic Regression

This project applies machine learning techniques to predict software bugs using the dataset Antfile17.CSV. The main goal is to classify whether a software module contains a bug (1) or not (0) based on 20 features.

Dataset

• File: Antfile17.CSV
• Features: 20 software metrics (columns)
• Label: bug (0 = No Bug, 1 = Bug)

---

Project Structure

🔹 KNN Classifier

Two approaches are shown (legacy sweep, plus improved pipeline + CV):

1. Legacy: sweep k = 1..30 and pick best on a simple split (kept for reference).
2. Improved: Pipeline (SMOTETomek → StandardScaler → KNN) with Stratified 5-fold GridSearchCV on train only, then evaluation on an untouched test set.

Updated workflow:

• Load dataset and check class distribution
• Train/test split: 80% train / 20% test (stratified)
• Imbalanced-learn Pipeline: SMOTETomek → StandardScaler → KNeighborsClassifier
• Hyperparameter tuning via GridSearchCV (CV on train only)
• Evaluate on test: Accuracy, F1, ROC AUC, Confusion Matrix, full classification report
• Diagnostics: Precision–Recall (Average Precision) and Calibration (Brier score + reliability curve)
• Persist best model: models/knn_pipeline.joblib

---

🔹 Logistic Regression

Pipeline (SMOTETomek → StandardScaler → LogisticRegression) with Stratified 5-fold GridSearchCV on train only.

Updated workflow:

• Load dataset
• Train/test split: 80% train / 20% test (stratified)
• Pipeline: SMOTETomek → StandardScaler → LogisticRegression(max_iter=1000)
• Hyperparameter tuning via GridSearchCV (CV on train only)
• Evaluate on test: Accuracy, F1, ROC AUC, Confusion Matrix, full classification report
• Diagnostics: Precision–Recall (Average Precision) and Calibration (Brier score + reliability curve)
• Persist best model: models/logreg_pipeline.joblib

---

Requirements

Install all required packages:

pip install scikit-learn
pip install imbalanced-learn
pip install seaborn
pip install matplotlib
pip install joblib

---

Model persistence and inference

Both notebooks save their best cross-validated pipelines under models/:

• models/knn_pipeline.joblib
• models/logreg_pipeline.joblib

Example: load a saved pipeline and predict on new data X_new (same columns/order as training features):

from joblib import load
import pandas as pd

# Load one of the persisted pipelines
pipe = load('models/logreg_pipeline.joblib')  # or 'models/knn_pipeline.joblib'

# X_new must be a DataFrame with the same feature columns as training
# X_new = pd.DataFrame([...], columns=[...])
pred = pipe.predict(X_new)
proba = pipe.predict_proba(X_new)[:, 1]
print(pred, proba)

Notes:

• The Pipeline encapsulates resampling and scaling; pass raw feature columns matching training schema.