Simulated Evolution and Learning 

Abstract: In this paper, a new genetic algorithm for multi-objective optimization problems is introduced. That is called ”Neighborhood Cultivation GA (NCGA)”. In the recent studies such as SPEA2 or NSGA-II, it is demonstrated that some mechanisms are important; the mechanisms of placement in an archive of the excellent solutions, sharing without parameters, assign of fitness, selection and reflection the archived solutions to the search population. NCGA includes not only these mechanisms but also the neighborhood crossover. The comparison of NCGA with SPEA2 and NSGA-II by some test functions shows that NCGA is a robust algorithm to find Pareto-optimum solutions. Through the comparison between the case of using neighborhood crossover and the case of using normal crossover in NCGA, the effect of neighborhood crossover is made clear.

Abstract: Anomaly detection aims to find patterns in data that are significantly different from what is defined as normal. One of the challenges of anomaly detection is the lack of labelled examples, especially for the anomalous classes. We describe a neural network based approach to detect anomalous instances using only examples of the normal class in training. In this work we train the net to build a model of the normal examples, which is then used to predict the class of previously unseen instances based on reconstruction error rate. The input to this network is also the desired output. We have tested the method on six benchmark data sets commonly used in the anomaly detection community. The results demonstrate that the proposed method is promising for anomaly detection. We achieve F-score of more than 90% on 3 data sets and outperform the original work of Hawkins et al. on the Wisconsin breast cancer set.

Abstract: Feature selection is an important pre-processing step in classification tasks. Feature selection aims to minimise both the classification error rate and the number of features, which are usually two conflicting objectives. This paper develops a differential evolution DE based multi-objective feature selection approach. The multi-objective approach is compared with two conventional methods and two DE based single objective methods, where the first algorithm is to minimise the classification error rate only while the second algorithm combines the number of features and the classification error rate into a single fitness function. Their performances are examined on nine different datasets and the results show that the proposed multi-objective algorithm successfully evolved a number of trade-off solutions, which reduce the number of features and keep or reduce the classification error rate. In almost all cases, the proposed multi-objective algorithm achieved better performance than all the other four methods in terms of both the classification accuracy and the number of features.

Abstract: Hyper-GA was introduced by the authors as a genetic algorithm based hyperheuristic which aims to evolve an ordering of low-level heuristics so as to find a good quality solution to a given problem. The adaptive length chromosome hyper-GA, let’s call it ALChyper-GA, is an extension of the authors previous work, in which the chromosome was of fixed length. The aim of a variable length chromosome is two fold; 1) it allows dynamic removal and insertion of heuristics 2) it allows the GA to find a good chromosome length which could otherwise only be found by experimentation. We apply the ALChyper-GA to a geographically distributed training staff and courses scheduling problem, and report that good quality solution can be found. We also present results for four versions of the ALChyper-GA, applied to five test data sets.