Table 12. ML-enhanced optimization algorithms for smart city applications.

Algorithm	Application domain	Strengths	Limitations	Main performance and representative cases
XGBoost + NSGA-II	Urban energy optimization	High efficiency; substitutes CFD models	Limited interpretability; sensitive to data quality	420 $\times$ faster than CFD (Liu et al., 2024a)
CNN-LSTM	Smart building HVAC	Handles temporal patterns; enhances energy use	High training cost; generalization challenges	22.3% energy savings; 28.5% comfort gain (Shafiq et al., 2024)
DRL	Traffic signal control	Adaptive; supports real-time decisions	Data-intensive; costly training	Reduced vehicle wait times; improved flow (Dong et al., 2024)
HNN + COOT	Air pollution forecasting	Accurate; supports multi-objective learning	Difficult to interpret; low transparency	47% MAE reduction (NO2, SO2) (Jalali et al., 2024)
Pix2Pix GAN + GA	Urban thermal design	Adapts to complex layouts; robust design support	Needs large, clean datasets	Improved ventilation and thermal comfort (Wu et al., 2024)