TY - JOUR
T1 - A modified multi-objective slime mould algorithm with orthogonal learning for numerical association rules mining
AU - Yacoubi, Salma
AU - Manita, Ghaith
AU - Amdouni, Hamida
AU - Mirjalili, Seyedali
AU - Korbaa, Ouajdi
N1 - Publisher Copyright:
© 2022, The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature.
PY - 2022
Y1 - 2022
N2 - Association rule mining (ARM) is defined by its crucial role in finding common pattern in data mining. It has different types such as fuzzy, binary, numerical. In this paper, we introduce a multi-objective orthogonal mould algorithm (MOOSMA) with numerical association rule mining (NARM) which is a different type of ARM. Existing algorithms that deal with the NARM problem can be classified into three categories: distribution, discretization and optimization. The proposed approach belongs to the optimization category which is considered as a better way to deal with the problem. Our main objective is based on four efficiency measures related to each association: Support, Confidence, Comprehensibility, Interestingness. To test the performance of our approach, we started by testing our method on widely known generalized dynamic benchmark tests called CEC’09. This benchmark is composed of 20 test functions: 10 functions without constraints and 10 functions with constraints. Secondly, we applied our algorithm to solve NARM problem using 10 frequently used real-world datasets. Experimental analysis shows that our algorithm MOOSMA has better results in terms of Average Support, Average Confidence, Average Lift, Average Certain factor and Average Netconf. Note that source code of the MOOSMA algorithm is publicly available at https://github.com/gaithmanita/MOOSMA.
AB - Association rule mining (ARM) is defined by its crucial role in finding common pattern in data mining. It has different types such as fuzzy, binary, numerical. In this paper, we introduce a multi-objective orthogonal mould algorithm (MOOSMA) with numerical association rule mining (NARM) which is a different type of ARM. Existing algorithms that deal with the NARM problem can be classified into three categories: distribution, discretization and optimization. The proposed approach belongs to the optimization category which is considered as a better way to deal with the problem. Our main objective is based on four efficiency measures related to each association: Support, Confidence, Comprehensibility, Interestingness. To test the performance of our approach, we started by testing our method on widely known generalized dynamic benchmark tests called CEC’09. This benchmark is composed of 20 test functions: 10 functions without constraints and 10 functions with constraints. Secondly, we applied our algorithm to solve NARM problem using 10 frequently used real-world datasets. Experimental analysis shows that our algorithm MOOSMA has better results in terms of Average Support, Average Confidence, Average Lift, Average Certain factor and Average Netconf. Note that source code of the MOOSMA algorithm is publicly available at https://github.com/gaithmanita/MOOSMA.
KW - Association rules mining
KW - Multi-objective optimization
KW - Orthogonal learning
KW - Swarm intelligence
UR - http://www.scopus.com/inward/record.url?scp=85141956074&partnerID=8YFLogxK
U2 - 10.1007/s00521-022-07985-w
DO - 10.1007/s00521-022-07985-w
M3 - Article
AN - SCOPUS:85141956074
SN - 0941-0643
JO - Neural Computing and Applications
JF - Neural Computing and Applications
ER -