Project-AI-and-Machine-Learning-uom-2025

📊 Bankruptcy Prediction using Machine Learning

Addresses the problem of predicting whether a company will go bankrupt based on financial ratios using several classification techniques.

---

📁 Structure

The repository includes:

• 📂 project2company/: All core Python scripts, data preprocessing, model training and evaluation.
• 📄 ΕφΠλη_Εργασία 2.pdf: Final report (Assignment 2).
• 📄 ΕφΠλη_Εργασία 3.pdf: Follow-up report.
• 📄 Καλές πρακτικές checklist.pdf: Report writing guidelines.

---

⚙️ Tools & Technologies

• Python 3.x
• Pandas, NumPy, Matplotlib, Seaborn
• Scikit-learn (MinMaxScaler, StratifiedKFold, classifiers)
• Google Colab (execution environment)
• Excel (for pivot table visualization)

---

🔍 Main Steps of the Pipeline

1. Data Validation – Check for missing values (NaN)
2. Normalization – Apply MinMax scaling to numeric features
3. Stratified K-Fold Split (k=4)
4. Downsampling – Balance classes (3:1 ratio of healthy to bankrupt)
5. Model Training & Evaluation on 8 classifiers:
   • Logistic Regression, LDA, KNN, Decision Tree, Random Forest, SVM, Naive Bayes, Gradient Boosting
6. Performance Metrics – Accuracy, Precision, Recall, F1, AUC, Recall_Healthy
7. Confusion Matrix – Train/Test visualized
8. Export to .csv for further Excel-based analysis

---

📊 Results & Visualization

• Output metrics are saved to: balancedDataOutcomes.csv
• Excel pivot tables were used to compare average performance across classifiers.
• Visual comparisons: stacked bar charts, F1 vs Recall, grouped charts.

---

▶️ How to Run

Google Colab (recommended)

1. Open any .py script from this repository in Google Colab
2. Run the notebook cells sequentially.
3. Outputs will appear inline (metrics, confusion matrices, graphs).

Alternatively, download the repo and run locally with:

pip install -r requirements.txt
python check_and_normalize.py
python model_loop_all.py