Email Spam Detection: A Symbiotic Feature Selection Approach Fostered by Evolutionary Computation

Abstract

The electronic mail (email) is nowadays an essential communication service being widely used by most Internet users. One of the main problems affecting this service is the proliferation of unsolicited messages (usually denoted by spam) which, despite the efforts made by the research community, still remains as an inherent problem affecting this Internet service. In this perspective, this work proposes and explores the concept of a novel symbiotic feature selection approach allowing the exchange of relevant features among distinct collaborating users, in order to improve the behavior of anti-spam filters. For such purpose, several Evolutionary Algorithms (EA) are explored as optimization engines able to enhance feature selection strategies within the anti-spam area. The proposed mechanisms are tested using a realistic incremental retraining evaluation procedure and resorting to a novel corpus based on the well-known Enron datasets mixed with recent spam data. The obtained results show that the proposed symbiotic approach is competitive also having the advantage of preserving end-users privacy.

A Survey on Evolutionary Computation Approaches to Feature Selection

“…For wrapper approaches, different classification algorithms have been used to evaluate the goodness of the selected features, e.g. SVMs [68], [71], [72], [73], [75], [79], [80], [81], [86], [107], KNN [39], [74], [76], [77], [80], [81], [86], [95], [107], ANNs [61], [69], [78], [81], [83], [85], DT [60], [80], [107], NB [80], [107], [109], multiple linear regression for classification [59], extreme learning machines (ELMs) [110], and discriminant analysis [66], [67], [82]. SVMs and KNN are the most popular classification algorithms due to their promising classification performance and simplicity, respectively.…”

“…Winkler et al [81] proposed several fitness functions, which considered the number of features, the overall classification performance, the class specific accuracy, and the classification accuracy using all the original features. Sousa et al [109] employed a fitness function using area under curve (AUC) of the receiver operating characteristic (ROC) of a NB classifier. In [107], a filter measure (Pearson correlation measure) and a wrapper measure (classification accuracy) were combined to form a single fitness function in a GA for feature selection to utilise the advantages of each measure.…”

“…Customer churn prediction [92] 4Text mining [16] 4Web service [141] 4Network security [97], [184], [220] 4Email Spam detection [109] 5Power system optimisation [79], [152] (5) Weed recognition in agriculture [191] (5) Melting point prediction in chemistry [188] (5) Weather forecast [86], [195] approaches is usually worse, but they can be much cheaper than wrapper approaches [234], which is critical in large datasets. Therefore, developing filter measures specifically according to the characteristics of an EC technique may significantly increase the efficiency and effectiveness, which offers an important future research direction.…”

A self-adaptive level-based learning artificial bee colony algorithm for feature selection on high-dimensional classification

“…The forerunner is Siedlecki and Sklansky who prove that genetic algorithm (GA) is a powerful tool for feature selection when the dimensionality of the given feature set is greater than 20 (Siedlecki and Sklansky 1993). After that, to further improve the performance of GA for feature selection, many different improvements have been proposed in search mechanisms (Demirekler and Haydar 1999;Jeong et al 2015;Wang et al 2020) and fitness function (Canuto and Nascimento 2012;Sousa et al 2013). A feature selection method (Derrac et al 2009) based on GA utilizes the cooperative coevolution (CC) framework (Potter and Jong 2000) but is not investigated in the large datasets.…”

A Survey on Evolutionary Computation Approaches to Feature Selection

“…Customer churn prediction [92] 4Text mining [16] 4Web service [141] 4Network security [97], [184], [220] 4Email Spam detection [109] 5Power system optimisation [79], [152] (5) Weed recognition in agriculture [191] (5) Melting point prediction in chemistry [188] (5) Weather forecast [86], [195] approaches is usually worse, but they can be much cheaper than wrapper approaches [234], which is critical in large datasets. Therefore, developing filter measures specifically according to the characteristics of an EC technique may significantly increase the efficiency and effectiveness, which offers an important future research direction.…”