IEEE Journal & Transactions Camera-Ready Visuals

Publication-Grade Performance Graphs & Visual Analytics

High-impact figures, loss convergence curves, PCMCI causal discovery matrices, ablation studies, and confidence interval forecasting models formatted for IEEE publication standards.

Fig. 1 • Model Optimization

Training & Validation Loss Convergence

MSE Loss over 50 Training Epochs (Neural-Causal vs Baselines)

Optim: Adam (lr=1e-3, decay=1e-5) Min Val Loss: 0.0084 (Epoch 42)

Fig. 2 • Comparative Evaluation

Model Accuracy & Error Metrics

Multivariate Metric Comparison (R², RMSE, MAE)

Proposed Model: R² = 0.946 +16.5% R² over Standard LSTM

Fig. 3 • Causal Discovery

PCMCI Inter-Sensor Causal Matrix

Momentary Conditional Independence (MCI) Strength at Lag τ=1

Significance Threshold: p < 0.01 Max MCI Link: CO → AQI (0.81)

Fig. 4 • Ablation Study

Architectural Component Contributions

Incremental R² Accuracy Gains by Layer Module

Baseline R²: 0.752 (Vanilla LSTM) Causal Prior Boost: +0.068 R²

Fig. 5 • Health Risk Forecasting

24-Hour SRI Forecast with 95% Confidence Bounds

Ground Truth vs Neural-Causal Ahead Predictions (Subclinical Risk Index)

Horizon: 24 Steps Ahead 95% CI Coverage: 98.2%

Fig. 6 • Exposure Fingerprint

Multivariate Profile Radar

Baseline vs Acute Exposure Peak

Parameters: 4 Active Channels Surge Ratio: 2.8x