Table 5.
Accuracies for the empirical datasets [%]
| Dataset | SVM | Jrip | J48 | Naïve Bayes | Average | Standard deviation |
|---|---|---|---|---|---|---|
|
Cypraeidae |
94.32 |
86.93 |
91.76 |
93.18 |
91.55 |
2.82 |
|
Drosophila |
98.28 |
94.83 |
91.38 |
96.55 |
95.26 |
2.55 |
|
Inga |
89.83 |
88.14 |
88.14 |
91.53 |
89.41 |
1.41 |
|
Bats |
100.00 |
100.00 |
98.15 |
100.00 |
99.54 |
0.80 |
|
Fishes |
95.50 |
90.09 |
92.79 |
97.30 |
93.92 |
2.73 |
|
Birds |
98.42 |
84.86 |
91.80 |
94.32 |
92.35 |
4.93 |
|
Fungi |
80.00 |
50.00 |
60.00 |
70.00 |
65.00 |
11.20 |
| Algae | 100.00 | 60.00 | 60.00 | 100.00 | 80.00 | 20.00 |
Results of the Weka supervised learning methods tested on empirical datasets show that SVM and Naïve Bayes outperform the other techniques in term of percentage of the correct species identification. The differences between SVM and the other algorithms result statistically significant (p-value ≤ 0.001), except for Naive Bayes (p-value > 0.05). The best performances are highlighted in bold for each dataset.