Table 9 Average prediction error among machine learning methods for high and low review helpfulness groups.
Box office revenue | Number of prediction error data used to compare | Random forests | Decision trees using boosting | k-nearest neighbor method | Discriminant analysis |
|---|---|---|---|---|---|
Week 1 | |||||
Low helpfulness (881)a | 30 | 0.120 | 0.131 | 0.152 | 0.179 |
High helpfulness (881) | 30 | 0.077 | 0.089 | 0.127 | 0.148 |
t-value (p-value) | 2.255 (0.023) | 1.969 (0.054) | 1.256 (0.214) | 1.331 (0.188) | |
Week 2 | |||||
Low helpfulness (505) | 34 | 0.233 | 0.250 | 0.235 | 0.287 |
High helpfulness (505) | 34 | 0.111 | 0.093 | 0.136 | 0.113 |
t-value (p-value) | 4.960 (0.000) | 6.290 (0.000) | 4.172 (0.000) | 5.919 (0.000) | |
Week 3 | |||||
Low helpfulness (368) | 37 | 0.345 | 0.381 | 0.355 | 0.309 |
High helpfulness (368) | 37 | 0.259 | 0.309 | 0.360 | 0.247 |
t-value (p-value) | 2.430 (0.018) | 1.879 (0.064) | 0.146 (0.884) | 1.800 (0.076) | |
