Aumo-Tracy/aisip-ml-portfolio

A five-script program that uses python to demonstrate how to solve African related problems # African Economic Crisis — Neural Network Analysis **Africa AI Hub | Week 6 Task** ## Dataset - **Source:** Kaggle — Africa Economic Banking and Systemic Crisis Data - **Size:** 1059 rows × 14 columns - **Countries:** 13 African countries (1860–2014) - **Target:** Banking Crisis — binary classification (crisis / no crisis) ## What This Notebook Covers 1. Data loading, cleaning and exploration (Pandas) 2. Feature engineering and train/test split 3. Scikit-learn baseline — Random Forest Classifier 4. Keras Sequential Neural Network - Input layer → Dense(64, ReLU) → Dense(1, Sigmoid) - Trained over 50 epochs - Loss and accuracy curves plotted 5. Model comparison: Random Forest vs Neural Network 6. Experiments: - Added extra hidden layer (64 → 32) - Increased neurons to 128 - Added Dropout(0.3) ## Results | Model | Accuracy | |---|---| | Random Forest | XX% | | Base Neural Network (64) | XX% | | NN + Extra Layer | XX% | | NN + 128 Neurons | XX% | | NN + Dropout 0.3 | XX% | ## Key Finding > Neural networks are not always the best tool. > For structured tabular data like this dataset, > Random Forest performed competitively with far less complexity. ## Tools Used | Tool | Purpose | |---|---| | Pandas | Data loading and cleaning | | NumPy | Numerical operations | | Scikit-learn | Baseline ML model | | TensorFlow/Keras | Neural network | | Matplotlib | Loss curves and charts | ## Files - `african_crisis_nn.ipynb` — Main notebook - `loss_curves.png` — Training loss and accuracy plots - `model_comparison.png` — All models compared ## Author Built as part of the Africa AI Hub Data Science Programme