Rapid Rule Compaction Strategies for Global Knowledge Discovery in a Supervised Learning Classifier System

Abstract

Michigan-style learning classifier systems have availed themselves as a promising modeling and data mining strategy for bioinformaticists seeking to connect predictive variables with disease phenotypes. The resulting 'model' learned by these algorithms is comprised of an entire population of rules, some of which will inevitably be redundant or poor predictors. Rule compaction is a post-processing strategy for consolidating this rule population with the goal of improving interpretation and knowledge discovery. However, existing rule compaction strategies tend to reduce overall rule population performance along with population size, especially in the context of noisy problem domains such as bioinformatics. In the present study we introduce and evaluate two new rule compaction strategies (QRC, PDRC) and a simple rule filtering method (QRF), and compare them to three existing methodologies. These new strategies are tuned to fit with a global approach to knowledge discovery in which less emphasis is placed on minimizing rule population size (to facilitate manual rule inspection) and more is placed on preserving performance. This work identified the strengths and weaknesses of each approach, suggesting PDRC to be the most balanced approach trading a minimal loss in testing accuracy for significant gains or consistency in all other performance statistics.

An Extended Michigan-Style Learning Classifier System for Flexible Supervised Learning, Classification, and Data Mining

“…Previously, we demonstrated that UCS yielded the most promising performance on these types of simulated datsets when compared to XCS, MCS, GALE and GAssist (LCS algorithms) [5,6]. Therefore we utilize the 'core' version of UCS used in [7,9,10] as the standard of comparison for ExSTraCS. Notice that in Table 1 the 'core' ExSTraCS p-values are from a comparison to 'core' UCS, while all other p-values correspond to comparisons between 'core' ExSTraCS and ExSTraCS with respective mechanisms activated.…”

“…Discrete attribute SNP datasets very similar to those used in [7,9,10] were simulated using GAMETES [18] with; architectures at maximum and minimum detection difficulty, heritabilities (i.e. the proportion of class variance that can be attributed to modeled attributes) of (0.1, 0.2, or 0.4), a minor allele frequency of 0.2, 20 attributes (only four of which were predictive and 16 were noise), sample sizes of (200, 400, 800, or 1600) and a heterogeneous mix ratio of either (50:50 or 75:25) (e.g.…”

“…Rule compaction utilizes the whole training dataset to consolidate the classifier population with the goal of improving interpretation and knowledge discovery. Comparisons in [10] suggested that simple Quick Rule Filtering (QRF) was both the fastest, and particularly was well suited to the theme of global knowledge discovery [12] where it is more important to preserve or improve performance than to minimize rule population size(useful for knowledge discovery by manual rule inspection) [10]. This study applies QRF.…”

“…ExSTraCS makes the six rule compaction strategies evaluated in [10] available to post-process the classifier population (see (C) in Figure 1). Rule compaction utilizes the whole training dataset to consolidate the classifier population with the goal of improving interpretation and knowledge discovery.…”

Continuous Endpoint Data Mining with ExSTraCS

“…ExSTraCS 2.0 has been expanded and adapted to better suit the needs of real-world supervised learning problems wherein effective and efficient classification, prediction, data mining, and/or knowledge discovery is the goal. The features of ExSTraCS 2.0 that most differentiate it from XCS [20] or UCS [2] include: a rule specificity limit to address scalability issues [18], built-in rapid expert knowledge (EK) generation algorithms [14], EK guided covering and mutation for efficient learning and scalability [15,18], attribute tracking and feedback for reusing useful attribute combinations and characterizing patterns of heterogeneity [12], built-in rule compaction strategies [9], and the consolidation of explore/exploit to perform both simultaneously [13]. For a detailed description of the ExSTraCS 2.0 algorithm see [18].…”