abdelazizabakartahir/first-neural-network-africa-

# First Neural Network - African Data Exploration ## 📌 Overview This project builds and trains a **neural network using TensorFlow/Keras** on African economic data (GDP per capita, population, year) to predict **life expectancy**. It compares deep learning performance with traditional machine learning models (Linear Regression, Random Forest) and experiments with different architectures. ## 👤 Author **Abdelaziz Abakar Tahir** - AISIP Cohort 1 (Africa AI Hub) --- ## 📊 Dataset - **Source:** Gapminder / World Bank - **Time period:** 1952 - 2007 - **Countries:** 54 African nations - **Features:** - `year` - Year of observation - `pop` - Population - `gdpPercap` - GDP per capita (USD) - **Target:** `lifeExp` - Life expectancy (years) - **Samples:** 576 rows --- ## 🧠 Neural Network Architecture (Base Model) | Layer | Type | Neurons | Activation | |-------|------|---------|------------| | Input | Dense | 64 | ReLU | | Output | Dense | 1 | Linear | - **Optimizer:** Adam (learning rate = 0.001) - **Loss Function:** Mean Squared Error (MSE) - **Metrics:** Mean Absolute Error (MAE) - **Epochs:** 50 - **Batch Size:** 32 - **Validation Split:** 20% --- ## 📈 Loss Curves *Left: Training vs Validation Loss (MSE) over 50 epochs* *Right: Training vs Validation MAE over 50 epochs* The curves show convergence with no significant overfitting, as both training and validation losses decrease together. --- ## 📊 Model Comparison Results | Model | MSE | R² Score | |-------|-----|----------| | Neural Network (64 neurons) | 18.45 | 0.85 | | Linear Regression | 22.31 | 0.82 | | **Random Forest** | **12.67** | **0.89** | **Key Finding:** Random Forest outperformed the neural network on this small tabular dataset (576 samples). --- ## 🧪 Experimentation Results | Model | Architecture | MSE | R² Score | |-------|--------------|-----|----------| | Base Model | 64 neurons | 18.45 | 0.85 | | + Hidden Layer + Dropout | 64 → 32 + Dropout(0.2) | 19.23 | 0.84 | | Increased Neurons | 1 …