Machine learning algorithm and neural network architecture for optimization of pharmaceutical and drug manufacturing industrial effluent treatment using activated carbon derived from breadfruit (Treculia africana)

Journal of Engineering and Applied Science

Table 5 Predicted optimum based on the ML performance and response surface design space

Contact time	Dosage	pH	Particle size	Actual observation	RSM prediction	Predicted DT	Predicted RF
Minute	mg/L		mm	mg/L	mg/L	mg/L	mg/L
10	2.5	6	0.60	42	42.42	42.57	41.61
10	*2.5*	8	*0.15*	41	*39.69*	41.55	41.41
50	0.5	6	0.30	54	54.46	52.65	52.95
10	2.5	8	0.30	42	40.38	42.88	42.69
10	0.5	6	0.85	57	56.99	45.77	45.72
10	2.5	10	0.60	42	41.16	42.89	42.80
50	0.5	8	0.60	58	57.21	57.93	58.08
50	0.5	10	0.85	59	62.12	58.48	58.45
10	*2.5*	6	*0.15*	38	*40.32*	*41.74*	*41.23*
50	0.5	8	1.00	59	59.08	57.96	58.04
10	2.5	10	0.15	41	39.06	41.73	41.47
30	0.5	6	0.85	51	52.04	51.58	51.76
30	0.5	8	0.15	48	47.20	48.06	48.20
50	0.5	6	0.85	58	57.03	57.28	57.27
10	0.5	10	0.30	45	44.61	44.74	44.75
50	2.5	6	0.15	45	45.41	46.11	46.18
50	0.5	8	1.00	59	59.08	46.19	45.93
30	2.5	10	0.60	46	45.19	46.19	41.74
10	2.5	6	0.15	40	37.76	41.74	40.23
30	0.5	8	1.00	54	54.32	52.73	52.87
10	2.5	10	0.15	34	39.06	52.74	41.74

ML predicted optimum, pH of 6, dosage of 2.5 mg/L, contact time of 10 min, and particle size of 0.15 mm.
RF reliability metrics, Adj-R² = 0.8532, MSE = 2.06, RMSE = 2.7, and Std dev = ± 1.01.
DT reliability metrics: Adj-R² = 0.5000, MSE = 3.88, RMSE = 5.02, and Std dev. = ± 2.05.
ML predicted efficiency = 90%, residual TDS concentration ≤ 40 mg/L, initial TDS concentration of 450 mg/L.
RSM validation metrics: Adj-R² = 0.9585, MSE = 0.00388, RMSE = 0.00627, Std dev = ± 0.54.
RSM predicted efficiency = 91%, residual TDS concentration ≤ 38 mg/L. Initial TDS concentration of 450 mg/L