A convolutional neural network model for forecasting equipment failures in telecommunication networks before they occur, enabling proactive maintenance scheduling and reducing unplanned downtime.
| Metric | Value |
|---|---|
| Test Accuracy | 93.6% |
| Architecture | CNN (3 conv layers) |
| Regularisation | Dropout |
Input → Conv2D → MaxPooling → Conv2D → MaxPooling → Conv2D → MaxPooling
→ Flatten → Dense(128) → Dropout(0.5) → Dense(num_classes, softmax)
Three convolutional layers with max-pooling, followed by fully connected layers with dropout regularisation to prevent overfitting.
Applied to improve generalisation on limited fault event data:
- Rescaling
- Rotation range
- Width/height shift
- Horizontal flip
- Shear and zoom range
- Training and validation accuracy/loss curves plotted across epochs
- Confusion matrix generated on held-out test set
- Final test accuracy: 93.6%
Python · TensorFlow · Keras · NumPy · Matplotlib · Scikit-learn
Equipment failures in telecom networks cause service disruptions and expensive emergency repairs. Predictive maintenance models allow operators to schedule interventions during planned maintenance windows, significantly reducing downtime and operational costs.