Automatic design of ant-miner mixed attributes for classification rule discovery

Abstract

The version in the Kent Academic Repository may differ from the final published version. Users are advised to check http://kar.kent.ac.uk for the status of the paper. Users should always cite the published version of record.

Archive-Based Pheromone Model for Discovering Regression Rules with Ant Colony Optimization

“…where ω jl is the weight of the highest rule that uses the value v i l for attribute i in the archive, u i l is the number of rules that use the value v i l for attribute i in the archive (u i l = 0 corresponds to the special case where v i l is not used by the rules in the archive), η is the number of values from t i that are not used in the archive (η = 0 corresponds to the special case where all values are used), and q is the same parameter used in Equation (5). The categorical sampling procedure allow an ant to consider two components when sampling a new value.…”

“…where ω j is the weight associated with the j-th rule in the archive calculated according to Equation (5). Let R i denote a new solution sampled by ant i around the chosen solution R j for continuous attribute a, the Gaussian probability density function (PDF) is given by…”

“…ACO MV replaces the graph-based pheromone model with an archivebased pheromone model to guide ants in mixed variables search space using different sampling strategies according to the variable type. Helal and Otero [4], [5] presented the first data mining approach, to the best of our knowledge, that used an archive-based pheromone model instead of a graph-based pheromone model for classification problem.…”

“…An automatic algorithm design approach for Ant-Miner MA+G [5] was created to overcome this problem, which combines the graph-based and archive-based pheromone models. This automatically configured algorithm outperformed the original cAnt-Miner algorithm to a significant level, and solved the problems Ant-Miner MA faced when dealing with a large number of attributes.…”

Data stream classification with ant colony optimisation

“…The discretisation step, either as a preprocessing or dynamic step, is a time‐consuming procedure requiring multiple passes through the data as it evaluates candidates threshold values. More recently, 27,28 used an archive‐based pheromone model to handle continuous attributes directly without a discretisation procedure—this leads to significant gains in terms of computational time, allowing the algorithms to handle larger data sets but showed limitations when there is an increased number of attributes. ACO classification algorithms also employ a pruning procedure, which removes irrelevant attribute‐value condition added due to the stochastic nature of the construction process.…”