H. Blockeel, S. Dzeroski, B. Kompare, S. Kramer, B. Pfahringer, and W. Van Laer. Experiments in predicting biodegradability. Applied Artificial Intelligence, 18(2):157--181, 2004.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....with only binary features, expressed in first order logic (FOL) see [Kramer et al. 2000] Kramer, 1999] Kramer et al. 1998] Alphonse and Toivonen, 1999] Dehaspe et al. 1999] To our knowledge, the use of other aggregates than existence has been lim ited. One example is given in [Deroski et al. 1999], which describes a propositionalisation step where numeric attributes were defined for counts of different substructures. It is our aim to analyse the applicability of a broader range of aggregates. With a growing availability of algorithms from the fields of ILP and Multi Relational Data Mining ....

....and Financial is essentially a snowflake schema. Many real world datasets described in the literature as ILP applications essentially have such a manageable structure. Moreover, results on these datasets frequently exhibit the extra condition of Wv 1. Some illustrative examples are given in [Deroski et at. 1999] and [Todorovski et at. 1999] example The following regression tree is taken (with kind permission from the authors) from ILP experiments with biodegradability [Deroski et al. 1999] which involved a dataset very similar to the Mutagenesis dataset. Careful examination of the variable structure ....

[Article contains additional citation context not shown here]

Deroski, S., Blockeel, H., Kompare, B., Kramer, S., Pfahringer, B., Van Laer, W., Experiments in Predicting Biodegradability, 31 32

....relationships (QSARS s) one of the tasks is to predict the half life time in water for aerobic aqueous biodegradation of a compound, given its chemical structure. Results on this domain with some ILP systems can be found in [Van Laer et al. 1997] preliminary results with ICL) and [Deroski et al. 1999] (more recent results) In this section some results related to ICL are given, including some unpublished ones. The data set consists of 328 examples. The biodegradation time has been discretized into 4 classes: fast (up to 7 days) moderate (1 to 4 weeks) slow (1 to 6 months) and resistant. The ....

....that sulphur slows down biodegradation. Another rule states that a compound is fast to degrade if it contains a benzene and a phenol group and is lighter that 170. The expert comments that in this case degradability is probably due to hydrolisis and photolysis. Together with the authors of [Deroski et al. 1999], we performed some additional experiments on the biodegradability data ( Blockeel et al. 2001] P1 has a count of the number of occurrences of each functional group (total of 31 at tributes) P2 has a count of all substructures of 2, 3 or 4 atoms that appear in at least 3 compounds (total of ....

S. Deroski, H. Blockeel, S. Kramer, B. Kompare, B. Pfahringer, and W. Van Laer. Experiments in predicting biodegradability. In S. Deroski and P. Flach, editors, Proceedings of the Ninth International Workshop on Inductive Logic Programming, volume 1634 of Lecture Notes in Artificial Intelligence, pages 80-91. Springer-Verlag, 1999.

....(a) and on the training set accuracy (b) for the mutagenesis data. 125 6.2 Influence of number of thresholds on the size of the theory (a) and on the induction time (b) for the mutagenesis data . 126 6. 3 A lattice of background information for the biodegradability data, as given in [Blockeel et al. 2001] . 128 6.4 An incomplete listing of the rules generated by ICL for detecting traffic problems . 134 6.5 Representation in Prolog (in models format) of the gene with ID 6.6 Representation in Prolog (in keys format) of the gene with ID 6.7 Part of the ....

....if it contains a benzene and a phenol group and is lighter that 170. The expert comments that in this case degradability is probably due to hydrolisis and photolysis. Together with the authors of [Deroski et al. 1999] we performed some additional experiments on the biodegradability data ([Blockeel et al. 2001]) P1 has a count of the number of occurrences of each functional group (total of 31 at tributes) P2 has a count of all substructures of 2, 3 or 4 atoms that appear in at least 3 compounds (total of 61 attributes) Both include logr and roweight. Global P 1 Global P2 Global R Global Pl P2 ....

[Article contains additional citation context not shown here]

H. Blockeel, S. Deroski, B. Kompare, S. Kramer, B. Pfahringer, and W. Van Laer. Experiments in predicting biodegradability, 2001. Unpublished.

.... differing interests, reasonable people may disagree on the interpretation of ambiguous or conflicting 1 Automated prediction of chemical properties, such as carcinogenicity, on the basis of chemical structure and analytic comparisons with other chemicals is an active area of research, e.g. [15, 73]. However this work has not looked at combining such different types of evidence for properties. 2 Subsequent epidemiological studies have provided statistically significant evidence for human nasal and other cancers from exposure to formaldehyde [82] 2 evidence. In the formaldehyde case, for ....

S. Dzeroski, H. Blockeel, B. Kompare, S. Kramer, B. Pfahringer, and W. Van Laer. Experiments in predicting biodegradability. In S. Dzeroski and P. Flach, editors, Inductive Logic Programming (ILP-99). Springer Verlag, Berlin, Germany, 1999.

....marginally better than Progol and Tilde for background 1. For Background 2, 3 and 4 however, the performance of ICL, Progol and Tilde are similar. Note that the accuracy increases as more background is added. Results on the biodegradability domain can be found in [68] preliminary results) and [34] (more recent results) The task is to predict the half life in water for aerobic aqueous biodegradation of a compound from its chemical structure. tumor the accuracy is 39.9 Sigma1.0 with a theory size of 302.8. These are similar to our results. The experiments in that paper used a beam of size 5 ....

....the accuracy is 39.9 Sigma1.0 with a theory size of 302.8. These are similar to our results. The experiments in that paper used a beam of size 5 instead of 20. The other settings are the same. 21 Table 9: Accuracies of machine learning systems predicting Biodegradability. Results are taken from [34]. We have left out the results of the regression systems. System Representation Accuracy Accuracy ( C4.5 P1 55.2 86.2 C4.5 P2 56.9 82.4 RIPPER P1 52.6 89.8 RIPPER P2 57.6 93.9 FFoil R1 53.0 88.7 ICL R1 55.7 92.6 SRT C P1 R1 55.0 90.0 Tilde C R1 51.0 88.6 Tilde C P1 R1 52.0 89.0 ....

[Article contains additional citation context not shown here]

S. Dzeroski, H. Blockeel, et al. Experiments in predicting biodegradability. In Proceedings of the Ninth International Workshop on Inductive Logic Programming. Springer-Verlag, 1999.

No context found.

S. Dzeroski, H. Blockeel, B. Kompare, S. Kramer, B. Pfahringer, and W. Van Laer. Experiments in predicting biodegradability. In Proc. ILP-99. Springer, 1999.

No context found.

Dzeroski, S., Blockeel, H., Kompare, B., Kramer, S., Pfahringer, B., Van Laer, W.: Experiments in Predicting Biodegradability, in: ILP-99: Proceedings Ninth International Workshop on Inductive Logic Programming, Springer, Berlin, 1999.

No context found.

S. Dzeroski, H. Blockeel, B. Kompare, S. Kramer, B. Pfahringer, W. Van Laer. Experiments in predicting biodegradability. in: Proceedings of the 9th International Workshop on Inductive Logic Programming (ILP-99), 80-91, Springer, 1999.

....Pos and Neg are the two classes and for each of them a DNF theory is learned and evaluated. Multi merges the 2 theories into a multi class theory. The results for Progol, Foil and Tilde have been taken from [8] Results on the biodegradability domain can be found in [69] preliminary results) and [35] (more recent results) The task is to predict the half life in water for aerobic aqueous biodegradation of a compound from its chemical structure. The biodegradation time has been discretized into 4 classes: fast, moderate, slow and resistant. The structure of a compound is represented by facts ....

....in water for aerobic aqueous biodegradation of a compound from its chemical structure. The biodegradation time has been discretized into 4 classes: fast, moderate, slow and resistant. The structure of a compound is represented by facts about atoms and bonds, much like in the mutagenesis domain. In [35] experiments on the relational data (denoted R1) and 2 propositional versions of the data (denoted P1 and P 2) has been performed with the propositional classi cation systems C4.5 and RIPPER [15] and the relational learners FFoil [60] SRT [46] ICL and Tilde. A short overview of the results can ....

[Article contains additional citation context not shown here]

S. Dzeroski, H. Blockeel, S. Kramer, B. Kompare, B. Pfahringer, and W. Van Laer. Experiments in predicting biodegradability. In S. Dzeroski and P. Flach, editors, Proceedings of the Ninth International Workshop on Inductive Logic Programming, volume 1634 of Lecture Notes in Arti cial Intelligence, pages 80-91. SpringerVerlag, 1999.

....in a second phase using a standard statistical technique that assumes cumulativity. Note that this technique of feature construction (using a classical ILP system to construct structural features, then use a propositional learner) has often been used recently, usually with good results, see e.g. [22, 8]. But even though the experimental results are quite good, an approach such as the above one raises the following questions: Progol, like most rule based approaches, yields a rule set of minimal size. From the point of view of feature construction, it yields a small number of features that ....

....In these approaches, however, more indirect heuristics are used to construct features. Finally, we mention that there are many applications in the field of (Q)SAR modelling where we think there are a lot of opportunities for cumulativity oriented ILP techniques, e.g. biodegradability prediction [8], carcinogenicity prediction [22] 8 Conclusions In this paper we have hypothesised that the assumption of different features having (or not having) cumulative effects on a certain target property constitutes an important part of the inductive bias of many machine learning approaches. We ....

S. Dzeroski, H. Blockeel, S. Kramer, B. Kompare, B. Pfahringer, and W. Van Laer. Experiments in predicting biodegradability. In S. Dzeroski and P. Flach, editors, Proceedings of the Ninth International Workshop on Inductive Logic Programming, volume 1634 of Lecture Notes in Artificial Intelligence, pages 80--91. Springer-Verlag, 1999.

No context found.

H. Blockeel, S. Dzeroski, B. Kompare, S. Kramer, B. Pfahringer, and W. Van Laer. Experiments in predicting biodegradability. Applied Artificial Intelligence, 18(2):157--181, 2004.

No context found.

S. Dzeroski, H. Blockeel, B. Kompare, S. Kramer, B. Pfahringer, and W. Van Laer. Experiments in predicting biodegradability. In S. Dzeroski and P. Flach, editors, Proc. of the Ninth International Workshop on Inductive Logic Programming, volume 1634 of LNCS, pages 80--91. Springer, 1999.

No context found.

Dzeroski S., Blockeel H., Kramer S., Kompare B., Pfahringer B., and Van Laer W.. Experiments in predicting biodegradability. Proceedings of the Ninth International Workshop on Inductive Logic Programming (S. Dzeroski and P. Flach, eds.), LNAI, vol. 1634, Springer, pp. 80-91, 1999.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC