Table 5.
Classification accuracy on different relational data sets
| Propositionalization | Learner | Carc. | IMDB | Mut188 | Mut42 | Trains | MovieLens |
|---|---|---|---|---|---|---|---|
| MajorityVote | 0.55 | 0.73 | 0.67 | 0.69 | 0.50 | 0.72 | |
| Aleph (Perovšek et al. 2015) | J48 | 0.55 | 0.73 | 0.60 | 0.69 | 0.55 | – |
| Aleph (Perovšek et al. 2015) | SVM | 0.55 | 0.73 | 0.60 | 0.69 | 0.70 | – |
| RSD (Perovšek et al. 2015) | J48 | 0.60 | 0.75 | 0.68 | 0.98 | 0.60 | – |
| RSD (Perovšek et al. 2015) | SVM | 0.56 | 0.73 | 0.71 | 0.69 | 0.80 | – |
| RelF (Perovšek et al. 2015) | J48 | 0.60 | 0.70 | 0.75 | 0.76 | 0.65 | – |
| RelF (Perovšek et al. 2015) | SVM | 0.56 | 0.73 | 0.69 | 0.76 | 0.80 | – |
| Wordification (Perovšek et al. 2015) | J48 | 0.62 | 0.82 | 0.67 | 0.98 | 0.50 | – |
| Wordification (Perovšek et al. 2015) | SVM | 0.61 | 0.73 | 0.82 | 0.79 | 0.50 | – |
| Aleph (replicated) | J48 | 0.55 | – | 0.80 | 0.76 | 0.70 | – |
| Aleph (replicated) | SVM | 0.55 | – | 0.80 | 0.79 | 0.60 | – |
| RSD (replicated) | J48 | 0.56 | 0.84 | 0.88 | 0.92 | 0.60 | – |
| RSD (replicated) | SVM | 0.60 | 0.82 | 0.89 | 0.84 | 0.80 | – |
| Wordification (replicated) | J48 | 0.47 | 0.85 | 0.91 | 0.88 | 0.90 | 0.60 |
| Wordification (replicated) | SVM | 0.39 | 0.80 | 0.83 | 0.33 | 0.50 | 0.72 |
| Treeliker | J48 | 0.58 | – | 0.77 | 0.81 | 0.50 | – |
| Treeliker | SVM | 0.60 | – | 0.90 | 0.80 | 0.70 | – |
| PropDRM | 0.63 | 0.73 | 0.91 | 0.86 | 0.70 | 0.72 | |
| PropStar | 0.66 | 0.74 | 0.92 | 0.90 | 0.80 | 0.74 |
The best score for each dataset is in bold
For the proposed methods, we report average performance over 5 runs. The runs, marked with—were unable to finish in 12 h