Table 5.

Computational results for comparing LSTM-Opt with other ML algorithms^*

		Train	Test	pred	timeCPX	time	timeimp	inf(%)	optgap(%)	time	timeimp	inf(%)	optgap(%)	time	timeimp	inf(%)	optgap(%)
c	f	T	T	(%)		LSTM-Opt				LR				RF
3	10000	90	360	50	86415	11.3	7675	0.0	0.8	42.7	2024	70.0	1.9	76.6	1129	50.0	2.3
				75		1.4	63189	10.0	2.0	0.6	146459	90.0	17.3	-	-	100.0	-
				85		0.6	139042	20.0	2.9	-	-	100.0	-	-	-	100.0	-
				90		0.5	159647	30.0	3.6	-	-	100.0	-	-	-	100.0	-
				95		0.4	193403	50.0	4.5	-	-	100.0	-	-	-	100.0	-
				100		-	-	100.0	-	-	-	100.0	-	-	-	100.0	-
5	10000	90	360	50	86763	20.4	4253	0.0	0.9	8.2	10552	60.0	1.4	13.4	6491	40.0	1.9
				75		2.5	34371	0.0	1.6	0.7	124883	80.0	25.8	0.6	145415	90.0	24.2
				85		0.8	102727	0.0	2.3	-	-	100.0	-	-	-	100.0	-
				90		0.7	118189	0.0	2.7	-	-	100.0	-	-	-	100.0	-
				95		0.5	190865	10.0	3.3	-	-	100.0	-	-	-	100.0	-
				100		0.2	416680	60.0	14.2	-	-	100.0	-	-	-	100.0	-
8	10000	90	360	50	7.5	2.6	3	0.0	1.6	3.2	2	40.0	0.5	3.1	2	30.0	1.0
				75		1.0	8	10.0	1.9	0.7	8	80.0	14.3	0.7	8	90.0	22.5
				85		0.7	11	10.0	2.2	0.4	14	90.0	111.1	-	-	100.0	-
				90		0.6	13	20.0	2.4	-	-	100.0	-	-	-	100.0	-
				95		0.5	16	20.0	2.8	-	-	100.0	-	-	-	100.0	-
				100		0.2	38	20.0	3.8	-	-	100.0	-	-	-	100.0	-

^*Experiments only include ten test instances due to long solution times