189.

[Cover] [Contents] [Index]

Page 268

Table 6.2 Confusion matrices obtained by using maximum likelihood, ICM algorithm, SA algorithm and MPM algorithm

No.	1	2	3	4	5	6	7	u.a%.
(a) Maximum likelihood
1	8738	6160	872	498	1740	1 327	83	45.00
2	692	12178	2382	414	630	742	118	70.98
3	582	10671	4886	681	1420	965	19	25.42
4	279	760	133	10088	4175	384	59	63.53
5	961	1629	298	2840	19670	988	221	73.93
6	302	2152	1181	689	1502	5875	158	49.54
7	314	488	22	168	1005	60	1118	35.21
p.a.%	73.63	35.78	49.99	65.60	65.26	56.81	62.95	62553
(b) ICM algorithm
1	9441	4096	428	448	2353	599	83	54.11
2	350	15331	1804	376	341	317	13	82.73
3	575	10332	6065	512	951	609	2	31.84
4	56	303	106	10614	2583	120	2	7500
5	1014	1729	215	2835	21714	1042	168	75.61
6	126	1528	1084	508	1836	7641	43	59.85
7	306	719	72	85	364	13	1465	48.45
p.a.%	79.55	45.04	62.05	69.02	72.04	73.89	82.49	72271
(c) SA algorithm
1	10139	3714	123	462	2547	407	871	58.01
2	122	16676	1089	317	425	282	14	88.12
3	468	9740	7125	553	886	548	1	36.88
4	48	264	119	10979	1418	185	2	84.36
5	858	1829	222	2593	22919	932	174	77.62
6	26	1264	1070	463	1769	7982	23	63.36
7	207	551	26	M	178	5	1475	60.13
p.a.%	85.43	48.99	72.90	71.39	76.04	77.19	83.05	77295
(d) MPM algorithm
1	9938	3730	221	4451	2415	421	76	57.60
2	315	16188	1207	382	310	268	12	86.65
3	417	10313	6891	564	996	350	1	35.28
4	44	252	129	11122	1429	158	2	84.67
5	911	1823	224	2368	23229	1032	187	78.02
6	38	1142	1095	477	1616	8105	19	64.88
7	205	590	7	14	147	7	1479	60.39
p.a.%	83.74	47.56	70.50	72.32	77.07	78.38	83.28	76952
(a) Total Accuracy: 55.20%; kappa coefficient: 0.460 (b) Total Accuracy: 63.78%; kappa coefficient: 0.561 (c) Total Accuracy: 68.21%; kappa coefficient: 0.614 (d) Total Accuracy: 67.91%; kappa coefficient: 0.610

in Figure 6.21). The MPM algorithm achieved an overall classification overall accuracy of 67.91% (kappa=0.614), which is comparable to the result obtained by the SA algorithm.

For the ICM, MPM and SA algorithms, only pair-site neighbourhood

[Cover] [Contents] [Index]