Generating Accurate Rule Sets Without Global Optimization

By Eibe Frank, Ian H. Witten. 1998

Subfields: Code Generation & Program Synthesis

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Code Generation & Program Synthesis

Citations in the corpus

Interpretable Machine Learning: Fundamental Principles and 10 Grand Challenges (2021, methodology): “Associative classification methods assemble decision lists or decision sets from a set of pre-mined rules, generally either by greedily adding rules to the model one by one, or simply including all “top-scoring” rules into a decision set, where each rule is scored separately according to a scoring f”
A Survey of Methods for Explaining Black Box Models (2018, background): “Indeed, a decision tree can be linearized into a set of decision rules with the if-then form [31, 90, 91]:”
Learning Certifiably Optimal Rule Lists (2017, background): “, early algorithms [9, 15, 29]) and associative classi cation [41].”
How to See Hidden Patterns in Metamaterials with Interpretable Machine Learning (2021, methodology): “…that using the shape-frequency features, rather than the original raw features, improves the accuracy of classiﬁers for most machine learning methods , especially tree-based methods such as boosted trees, but with the exception of MLP (SFF decreases its in [0, 10] kHz, [10, 20] kHz, and [20, 30]”
Boolean Decision Rules via Column Generation (2018, background): “Predominant practical approaches include a covering or separate-and-conquer strategy ([15, 14, 16, 26, 28, 40],”
How do Humans Understand Explanations from Machine Learning Systems? An Evaluation of the Human-Interpretability of Explanation (2018, neutral): “Decision sets (also known as rule sets) also enjoy a long history of optimization techniques, ranging from Frank and Witten [1998], Cohen [1995], Clark and Boswell [1991] to Lakkaraju et al. [2016].”
Finding a short and accurate decision rule in disjunctive normal form by exhaustive search (2010, methodology): “• PART (Frank and Witten 1998): combines the decision tree approach and sequential covering by repeatedly generating partial decision trees.”
Machine learning: a review of classification and combining techniques (2006, methodology): “Once a partial tree has been build, a single rule is extracted from it and for this reason the PART algorithm avoids postprocessing (Frank and Witten 1998).”
A soft computing-based approach for integrated training and rule extraction from artificial neural networks: DIFACONN-miner (2010, unknown): “Dataset NBTree DecisionTable PART C4.5 DIFACONN-miner ECG Minimum 57.69 48 60 60 96 Average 67.45 65.48 71.39 70.17 97.45 Maximum 80.77 76 80.77 80.77 100 Standard deviation 6.88 8.97 6.62 7.08 0.021 Average no. of rules 2.3 3.3 4 3.2 8.45 Heart-C Minimum 35.484 50 38.71 41.935 90 Average 53.183 54.”
Directional decision lists (2015, methodology): “We end with an experiment on real data from a manufacturing process and determine indicators for problem cases.”
A Bayesian Framework for Learning Rule Sets for Interpretable Classification (2017, methodology): “…Chen et al., 2006; Cheng et al., 2007; McCormick et al., 2012; Rudin et al., 2013; Dong et al., 1999; Michalski, 1969; Clark and Niblett, 1989; Frank and Witten, 1998) mine for frequent rules in the data and combine them in a heuristic way, where rules are ranked by an interestingness…”
pre: An R Package for Fitting Prediction Rule Ensembles (2017, methodology): “Following Frank and Witten (1998), we distinguish between two strategies for generating rules: Indirectly, through transforming the nodes of one or more decision trees to a set of rules, and directly, through for example a sequential covering approach (F¨urnkranz 1999).”
Real-Time Detection of Traffic From Twitter Stream Analysis (2015, methodology): “The PART algorithm [48] combines two rule generation methods, i.e., RIPPER [51] and C4.5 [46].”
Improving the interpretability of classification rules discovered by an ant colony algorithm (2013, methodology): “The Unordered c Ant-Miner PB using the rule selection con-ﬂict resolution strategy (denoted as U-c AM PB [S]) achieved the best average rank, outperforming state-of-the-art rule induction algorithms with statistically signiﬁcant diﬀerences, namely PART, Unordered CN2, C4.5rules and JRip.”
SMOTE-IPF: Addressing the noisy and borderline examples problem in imbalanced classification by a re-sampling method with filtering (2015, methodology): “These are k-NN [39], a Support Vector Machine (SVM) [13] [51], Repeated Incremental Pruning to Produce Error Reduction (RIPPER) [12] and PART [16].”