Logistic Regression Demo

AIO2025: Module 06.

📊

About this Logistic Regression Demo

Interactive demonstration of Logistic Regression using NumPy and gradient descent. Learn binary classification with sigmoid activation, binary cross-entropy loss, and adjustable prediction threshold. Visualize training metrics and experiment with different threshold values.

📊 How to Use: Select binary classification data → Configure target (must have 2 classes) → Set training parameters → Adjust threshold → Enter feature values → Run training!

Start with sample datasets or upload your own CSV/Excel files.

📁 Upload Your Data

🗂️ Sample Datasets

🎯 Target Column

🔄 Loading sample data...

📋 Data Preview (First 5 Rows)

📊 Logistic Regression Parameters

Number of Epochs

Number of training iterations

Learning Rate (Power of 10)

0=1e-6, 1=1e-5, 2=1e-4, 3=1e-3, 4=1e-2, 5=1e-1, 6=1

0 6

Current Learning Rate: 0.01

Batch Size (Power of 2)

Slide to select: 0=1, 1=2, 2=4, 3=8, ... Max=Full Batch

0 10

Current Batch Size: Full Batch

📊 Data Split Configuration

Train/Validation Split Ratio

Proportion of data used for training (e.g., 0.8 = 80% train, 20% validation)

0.6 0.9

🎯 Prediction Threshold Configuration

Classification Threshold

Probability threshold for binary classification. Predict class 1 if probability ≥ threshold, else class 0. Adjust to balance precision/recall.

0 1

Current Threshold: 0.50

📊 Logistic Regression Results & Visualization

Training Loss & Accuracy Over Epochs

Validation Loss & Accuracy Over Epochs

**📊 Logistic Regression Results**

Training details will appear here showing model performance, learned parameters, and predictions with current threshold.

📊 Logistic Regression Guide:

📈 Training Metrics:

Loss (BCE): Binary Cross-Entropy loss decreases as model learns. Lower loss indicates better fit.
Accuracy: Classification accuracy improves during training. Monitor both training and validation accuracy.

🔧 Training Parameters:

Epochs: Number of complete passes through training data. More epochs = better learning, but watch for overfitting.
Learning Rate: Step size for gradient descent. Recommended: 0.001 to 0.01. Too high may cause instability.
Batch Size: Samples processed before updating parameters. Powers of 2: 1, 2, 4, 8... or Full Batch. Smaller = faster updates but noisier. Larger = more stable.
Train/Validation Split: Proportion of data for training vs validation. Default 80/20 split.

🎯 Threshold Parameter:

Threshold: Probability cutoff for binary classification. If predicted probability ≥ threshold → class 1, else → class 0.
Default: 0.5 (balanced)
Lower threshold (e.g., 0.3): More predictions of class 1 → higher recall, lower precision
Higher threshold (e.g., 0.7): Fewer predictions of class 1 → higher precision, lower recall
Experiment: Adjust threshold to see how predictions and accuracy change!

🧮 Algorithm Details:

Sigmoid Activation: Maps linear output to probability (0-1 range)
Binary Cross-Entropy Loss: Optimized for binary classification tasks
Feature Normalization: Automatic standardization (zero mean, unit variance) for stable training

💡 Tips:

Start with default parameters (100 epochs, learning rate 0.01, threshold 0.5)
Monitor validation metrics to detect overfitting
Adjust threshold based on your classification goals (precision vs recall)
Use batch size = Full Batch for most stable training