Predictability-based collective class association rule mining

Song, Kiburm; Lee, Kichun

doi:10.1016/j.eswa.2017.02.024

Cited by 41 publications

(15 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…K. Song et al [27] investigate the predictive ability of candidate rules for the first time and design a novel AC mining approach, called predictability-based collective class association rule (PCAR), to construct a classifier with high predictability by properly using cross-validation and aggregating the final rules. Theoretical analysis and experiments show their PCAR method only preserves the candidate rules with high predictability and removes the useless, redundant and low predictable rules simultaneously.…”

Section: Related Workmentioning

confidence: 99%

A Novel Approach for Associative Classification Based on Information Entropy of Frequent Attribute Set

Cao

Zhong

et al. 2020

IEEE Access

View full text Add to dashboard Cite

Associative Classification (AC) is envisioned as one of the most attractive classification approaches to facilitate managers for prediction and decision making in a highly accurate and easily interpretable way. The most of existing AC algorithms mainly focus on the two static metrics of association rules: support and confidence. However, in this paper, we identify a potential limitation of the confidence and further point out these AC algorithms merely consider the explicit expression of classification information and knowledge of Class Association Rules (CARs) and neglect the implicit expression of classification information and knowledge of CARs. And even the explicit expression of classification information and knowledge of CARs is equivalent, the implicit expression of classification information and knowledge of them may be different. Thus, these CARs will cause different influence on prediction or decision-making in terms of the interpretability and rationality of classification, and will result in predictive bias or decision-making bias in practice. In response to this drawback, we introduce the notion of information entropy and propose an innovative approach for associative classification based on information entropy of frequent attribute set, named EAC. Different from the existing schemes, the proposed EAC algorithm enjoys the following promising merits: (1) lower predictive bias or decision-making bias by levering the information entropy of frequent attribute set; (2) better interpretability and rationality by setting higher support threshold; (3)setting global optimum parameter dynamically through repeated trials with reducing the dimension of data sets. Experiments on 20 well-known benchmark data sets demonstrate that our EAC approach is highly competitive to other state-of-the-art AC algorithms in terms of predictive bias or decision-making bias, interpretability, and efficiency, which can be used to construct a new classifier efficiently and effectively in many realistic scenarios. INDEX TERMS Data mining, associative classification, information entropy, frequent attribute set.

show abstract

Section: Related Workmentioning

confidence: 99%

A Novel Approach for Associative Classification Based on Information Entropy of Frequent Attribute Set

Cao

Zhong

et al. 2020

IEEE Access

View full text Add to dashboard Cite

show abstract

“…Mlakar et al (2017) presented a single-objective binary cuckoo search using a novel individual representation. Other works can be found in (Pears & Koh, 2011;Song & Lee, 2017;Anuradha & Kumar, 2017;Feng et al, 2016;Huang et al, 2017;Vo, Pham, Le & Deng, 2017;Kieu, Vo, Le, Deng & Le, 2017).…”

Section: Related Workmentioning

confidence: 99%

ARM–AMO: An efficient association rule mining algorithm based on animal migration optimization

Sơn

Chiclana

Kumar³

et al. 2018

Knowledge-Based Systems

View full text Add to dashboard Cite

“…After obtain all the FPs (initial candidate rule set), we can measure the prediction power of a frequent pattern by calculating the prediction rate. First, we divide the training dataset into k equal-length folds along base station dimension and produce k loops using k equal-length folds [10], in each loop i, we create inner training set TRi with (k-1) folds and inner testing set TEi with 1 fold. For each candidate pattern P, we calculate the ratio pi of the P's support in TRi to patterns' amount of same size in TRi and the ratio qi of the Q's (P's prefix-path) support in TEi to patterns' amount of same size in TEi.…”

Section: Several Criteria Of Rule Evaluationmentioning

confidence: 99%

“…Hernndez-Len [9] introduced the use of Netconf measure instead of support and confidence, then proposed an approach to rule ranking by combining Netconf with rules of a long length. Kiburm Song [10] applied cross-validation and aggregating rules to select ARs, defined a new concept of predict rate to measure algorithm prediction performance, removing redundant and low predictive power rules. In this paper, we propose a new comprehensive evaluation criterion of association rules including the predict rate to select the reliable rule.…”

Section: Introductionmentioning

confidence: 99%

A Novel Selection and Matching of Patterns Model for Traffic Prediction in Wireless Network

Shen¹,

Gao²

2018

dtcse

View full text Add to dashboard Cite

Data mining techniques are becoming increasing popular and important in the big data era. Among them, frequent patterns mining technique has matured, with this technique, we can obtain the association rules to predict the data in cellular network. Existing research mainly focus on the frequent item sets mining, rarely involves in the time series and FPM algorithm usually used user-defined minimum support in various training datasets, which is rigorous, may influence the number of the frequent patterns and fail to select the reasonable patterns. Moreover, traditional association rules mining only relies on single evaluation criterion such as confidence or support, which leads to select unsound rules. This paper introduces an optimized frequent pattern mining algorithm, improving the setting of support in the process of mining frequent sequences and presenting a new evaluation criterion for the candidate patterns. Experiments were conducted to select appropriate parameters, support and evaluation criterion. Furthermore, we apply the above conclusion to the cellular flow data prediction, and compare the runtime, matching rate, RMSE and MAPE of proposed algorithms with those of improved Markov algorithm to examine the effectiveness of algorithm.

show abstract

Predictability-based collective class association rule mining

Cited by 41 publications

References 12 publications

A Novel Approach for Associative Classification Based on Information Entropy of Frequent Attribute Set

A Novel Approach for Associative Classification Based on Information Entropy of Frequent Attribute Set

ARM–AMO: An efficient association rule mining algorithm based on animal migration optimization

A Novel Selection and Matching of Patterns Model for Traffic Prediction in Wireless Network

Contact Info

Product

Resources

About