Learning Classifier Systems

Butz, Martin V.

doi:10.1007/978-3-662-43505-2_47

Cited by 13 publications

(8 citation statements)

References 52 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While the principles of anticipatory learning modeling have been studied for several decades [28,130], IoMT is actually in its infancy. Although recently, researchers attempted to integrate an anticipatory process into artificial learning systems [131][132][133][134][135], few attempts can be found on research applications that apply the theory of anticipatory computing to building context intelligence in IoMT devices [136,137]. We advocate that the proliferation of IoMT devices has created a unique opportunity to explore anticipatory learning models using the vast amount of IoMT data streams.…”

Section: Research Challenges and Opportunitiesmentioning

confidence: 99%

A Holistic Overview of Anticipatory Learning for the Internet of Moving Things: Research Challenges and Opportunities

Cao

2020

IJGI

View full text Add to dashboard Cite

The proliferation of Internet of Things (IoT) systems has received much attention from the research community, and it has brought many innovations to smart cities, particularly through the Internet of Moving Things (IoMT). The dynamic geographic distribution of IoMT devices enables the devices to sense themselves and their surroundings on multiple spatio-temporal scales, interact with each other across a vast geographical area, and perform automated analytical tasks everywhere and anytime. Currently, most of the geospatial applications of IoMT systems are developed for abnormal detection and control monitoring. However, it is expected that, in the near future, optimization and prediction tasks will have a larger impact on the way citizens interact with smart cities. This paper examines the state of the art of IoMT systems and discusses their crucial role in supporting anticipatory learning. The maximum potential of IoMT systems in future smart cities can be fully exploited in terms of proactive decision making and decision delivery via an anticipatory action/feedback loop. We also examine the challenges and opportunities of anticipatory learning for IoMT systems in contrast to GIS. The holistic overview provided in this paper highlights the guidelines and directions for future research on this emerging topic.

show abstract

Section: Research Challenges and Opportunitiesmentioning

confidence: 99%

A Holistic Overview of Anticipatory Learning for the Internet of Moving Things: Research Challenges and Opportunities

Cao

2020

IJGI

View full text Add to dashboard Cite

show abstract

“…There are two classes of LCS's: Pittsburgh and Michigan. The difference between the two, lies in the way they evolve the classifiers: in Pittsburgh approach, each chromosome is a candidate rule set while in Michigan approach, each chromosome is a single rule and the whole population forms the classifier rule set [3].…”

Section: Introductionmentioning

confidence: 99%

An XCS-Based Algorithm for Classifying Imbalanced Datasets

Sanatkar

Haratizadeh

2015

IJIIS

View full text Add to dashboard Cite

Imbalanced datasets are datasets with different samples distribution in which the distribution of samples in one class is scientifically more than other class samples. Learning a classification model for such imbalanced data has been shown to be a tricky task. In this paper we will focus on learning classifier systems, and will suggest a new XCS-based approach for learning classification models from imbalanced data sets. The main idea behind the suggested approach is to update the important parameters of the learning method based on the information gathered in each step of learning, in order to provide a fair situation for the minor class, to contribute in building the final model. We have also evaluated our approach by testing it with real-world known imbalanced datasets. The results show that our new algorithm has a high detection rate and a low false positive rate.

show abstract

“…When the Carry problem scales by 2 bits the covered sub-space will increase by one, e.g. in 8-bits [2,4,24,8], on sub-space 2-5, 10-bit [2,4,8,48,16], and 12-bit [2,4,8,16,96,32]. Hence, the Carry problems' optimal rule set size = 4).…”

Section: Problemmentioning

confidence: 99%

“…Learning Classifier Systems (LCSs) have been identified as useful data mining techniques that can obtain optimal results that contain humandiscernable patterns [76] for a number of tasks. Previously, Butz described LCSs are designed to evolve a minimal number of non-overlapping rules to represent an explored domain, accurately and completely [16]. Such optimal rulesets term as [O] sets.…”

Section: Chapter 7 Absumption and Subsumption Based Lcs (Ascs)mentioning

confidence: 99%

Learning Classifier Systems for Understanding Patterns in Data

Liu¹

View full text Add to dashboard Cite

In the field of data-mining, symbolic techniques have produced optimal solutions, which are expected to contain informative patterns. Visualizing these patterns can improve the understanding of the ground truth of the explored domain. However, up to now, the symbolic algorithms struggle to produce optimal solutions for domains that have an overlapping distribution of feature patterns. Furthermore, the majority of problems have an overlapping distribution. Thus, novel techniques are needed to improve symbolic techniques’ capacity to address overlapping domains, so that it is practicable to achieve the visualization of the underlying patterns. Michigan-style Learning Classifier Systems (LCSs) are rule-based symbolic learning systems that utilize evolutionary computation to construct a population of rules to capture patterns and knowledge of the explored domains. LCSs have been applied to many data-mining tasks and have achieved good performance. Recently, employing visualization methods to improve the understanding level of models has become more and more popular in the data-mining community. In the LCSs field, visualization techniques orientated to explore feature patterns have not been developed. Investigating LCSs’ models is commonly based on reading rules or viewing their distribution. However, LCSs’ models may contain hundreds or even thousands of unduplicated rules, which makes the identification of patterns challenging. Previously, Butz defined LCSs’ optimal solutions as [O] sets, which are expected to utilize a minimal number of non-overlapping rules to present an explored domain completely and correctly. In the last two decades, rule compaction algorithms have been designed to search for [O]s by compacting LCSs’ models, where rules that violate [O]’s definition are considered redundant rules. However, in many problems, an ideal [O] does not exist. Even if such a ruleset exists, redundant rules are often discovered in the compacted models. Furthermore, compaction often results in a decreased prediction performance. The LCSs community used to believe the reduced performance is an unavoidable and acceptable price for producing a compacted model. It is observed that across multiple LCS produced populations for the same problem, the irrelevant/redundant rules are varied, but useful/accurate rules are consistent. According to this observation, this thesis collects the common accurate rules and finds that for an arbitrary clean dataset, the common rules can form a determinate and unique solution, i.e. the proposed natural solution. A natural solution is composed of all consistent and unsubsumable rules under the global search space. A natural solution can correctly and completely represent the explored clean datasets. Furthermore, searching for natural solutions can produce concise correct solutions without reducing performance. To visualize the knowledge in the solutions, three visualization methods are developed, i.e. Feature Important Map (FIM), Action-based Feature Importance Map (AFIM), and Action-based Average value Map (AFVM). FIM can trace how LCSs form patterns during the training process. Besides, AFIM and AFVM precisely present patterns in LCSs’ optimal solution respectively regarding attribute importance and specified attribute’s value’s importance. For the sake of efficiently producing natural solutions, a new type of compaction algorithm is introduced, termed Razor Cluster Razor (RCR). RCR is the first algorithm that considers Pittsburgh-style LCSs’ conceptions to Michigan-style LCSs’ compaction algorithms, i.e. compacting is based on multiple models. RCR was first designed for Boolean domains, then RCR has been extended to adapt to real-value LCSs’ models. The conducted experiments demonstrated that natural solutions are producible for both Boolean domains and real domains. The experiments regarding producing natural solutions lead this thesis to discover that LCSs have an over-general issue when addressing domains that have an overlapping distribution, i.e. optimal rules’ encodings naturally share overlapped niches. The over-general issue causes LCSs to fail in maintaining the prediction performance, i.e. due to the optimal rules and over-general rules being repeatedly introduced and removed, training performance can never achieve 100% accuracy. This newly discovered issue inspires the development of the Absumption method as a complement to the existing Subsumption method, which continuously seeks to produce optimal rules by correcting over-general rules. Absumption not only improves LCSs’ prediction performance in overlapping domains but also enables LCSs to employ hundreds of cooperative rules to precisely represent an explored domain. The success of Absumption demonstrates LCSs’ capacity in addressing overlapping domains. However, introducing Absumption to LCSs does increase the cost of computer resources for training, which results in a need for more efficient exploration. Furthermore, LCSs employed search techniques tend to evolve maximally generalized rules, rather than produce rules that do not overlap with the existing rules in the population. Thus, [O]s do not fit LCSs’ fundamental search techniques. As a result, LCSs’ accurate models often do not contain an optimal solution, which results in the LCSs produced models being poorly interpretable. This hampers LCSs from being a good data-mining technique. In this thesis, the Absumption and Subsumption based Learning Classifier System (ASCS) is developed. ASCSs consider natural solution as the search objective and promote the Absumption mechanism and Subsumption mechanism as the primary search strategies. Thus, it is possible to remove the traditional evolutionary search algorithms, i.e. crossover, mutation, roulette wheel deletion, and competition selection. Experiments demonstrated that ASCSs can use thousands of cooperative rules to represent an explored domain and enable easy pattern visualization that was previously not possible.

show abstract

Learning Classifier Systems

Cited by 13 publications

References 52 publications

A Holistic Overview of Anticipatory Learning for the Internet of Moving Things: Research Challenges and Opportunities

A Holistic Overview of Anticipatory Learning for the Internet of Moving Things: Research Challenges and Opportunities

An XCS-Based Algorithm for Classifying Imbalanced Datasets

Learning Classifier Systems for Understanding Patterns in Data

Contact Info

Product

Resources

About