Research Article
Deep Learning Approaches for Cyberbullying Detection and Classification on Social Media
Table 1
Consequences of existing with future feature assortment methods on various training data size.
| Classification accuracy on 60% training data | Residuals | BCO-FSS | Pearson correlation | Chi-squared | Information gain |
| 200 | 28.61 | 27.10 | 25.18 | 23.21 | 400 | 31.87 | 30.47 | 28.26 | 26.54 | 600 | 33.09 | 31.23 | 29.43 | 27.41 | 800 | 35.30 | 33.15 | 31.29 | 28.97 | 1000 | 37.80 | 34.78 | 32.68 | 29.46 |
| Classification accuracy on 75% training data | Residuals | BCO-FSS | Pearson correlation | Chi-squared | Information gain |
| 200 | 43.24 | 40.39 | 36.26 | 34.65 | 400 | 53.24 | 50.07 | 44.36 | 41.84 | 600 | 55.94 | 51.66 | 47.85 | 43.75 | 800 | 61.02 | 56.10 | 51.66 | 48.06 | 1000 | 66.26 | 60.07 | 55.47 | 52.19 |
| Classification accuracy on 90% training data | Residuals | BCO-FSS | Pearson correlation | Chi-squared | Information gain |
| 200 | 47.44 | 43.88 | 39.55 | 33.29 | 400 | 61.45 | 56.10 | 49.23 | 44.62 | 600 | 66.80 | 57.88 | 45.92 | 40.90 | 800 | 67.82 | 60.69 | 52.28 | 48.47 | 1000 | 74.18 | 69.60 | 64.51 | 59.61 |
|
|