Hybridizing Machine and Deep Learning for Urban Water Demand Forecasting: An Ensemble Framework Leveraging Dam Monitoring Data

Table 3 Comprehensive performance comparison of all models across training, validation, and test sets

Model	Dataset	R²	Rank (R²)	RMSE (m³/day)	Rank (RMSE)	MAE (m³/day)	Rank (MAE)	Training Time(s)
Random Forest	Train	0.9513	4	53,151	5	36,835	5	218
	Validation	0.7800	6	78,050	6	60,097	6
	Test	0.5193	6	133,323	6	96,029	6
XGBoost	Train	0.9336	5	62,046	6	44,485	6	154
	Validation	0.7826	5	77,588	5	59,385	5
	Test	0.4894	7	137,406	7	97,810	7
LightGBM	Train	0.9527	1	52,386	1	36,958	1	92
	Validation	0.7856	4	77,047	4	58,961	4
	Test	0.4729	5	139,612	5	99,702	5
LSTM	Train	0.9321	6	62,762	7	44,014	3	1120
	Validation	0.8304	3	68,520	3	51,868	3
	Test	0.8345	2	78,237	3	59,059	3
SVR	Train	0.9339	2	61,891	2	41,327	1	890
	Validation	0.8416	1	66,231	1	47,668	1
	Test	0.8566	1	72,815	1	51,762	1
Linear Model	Train	0.9241	7	66,340	3	47,005	2	5
	Validation	0.8240	2	69,805	2	51,077	2
	Test	0.8120	3	83,387	4	62,643	4
Ensemble	Test	0.8469	2	75,244	2	55,725	2	–