SOUTHERN RAILWAY // PROBLEM STATEMENT 26027

Neural Network Architecture of BlockTrain

R² = 93.21% Accuracy = 87.66% 100% Emergency Precision
2. Safety Risk 8.5 / 10
3. Overdue Days 21 Days
4. Asset Age 7.5 Years
Real-Time Network Decision CRITICAL EMERGENCY MPI: 68.4 / 100
INPUT: Multi-Dept Defect Normalized Tensor 1 OUTPUT "CRITICAL" Shadow Block 01:00 AM Layer 1 Raw Defect Values Detected L2 Interactions Identified L3 Combinations of MPI Regressed L4 Urgency Tiers Identified L5 Combinations of Shadow Actuation
Layer 1: Sensors Ingests raw physics: Safety Risk ($1-10$), Overdue Days, Asset Age, Repair Hours, Station Chainage Km, and Dept one-hots.
L2: Interactions Computes non-linear terms: $5.0 \times \text{Risk} + 0.7 \times \text{Overdue}$ and $\text{Risk} \times \text{Age}$ to model accelerating physical fatigue.
L3: Stage 1 MPI 150 boosted regression trees assemble the continuous Maintenance Priority Index (MPI: 0–100) ($R^2 = 93.21\%$).
L4: Safety Classifier Two-stage model stacking fuses raw features with MPI. 480 decision trees classify 4 tiers with 100% Emergency Precision.
L5: Actuation Head Groups multi-department possessions ($\pm 2.5\text{ km}$), cascades signals to Red Danger, and enforces pneumatic train braking.