Article Preview
Top1. Introduction
Information and communication technology and usage of the internet is the prime factor for the massive amount of data generated each day. These accumulated data is useless unless being analyzed to get some meaningful information. Processing of these data to get meaningful information is data analytics. Additionally identifying features affecting decisions is of challenging. Machine learning techniques are used to customize and supervise data to discover useful knowledge, rules and chief factors. The extracted knowledge and rules must be accurate, readable, comprehensible, and ease of understanding. Feature selection is a primary issue in machine learning, data mining and pattern recognition to preserve the meaning of information about a given problem. Additionally, it is imperative and aims to find the minimal subset of the original features. Simultaneously, it is a crucial pre-processing technique useful for data analysis, where only a subset from the original data features is chosen by to eliminate noisy, irrelevant or redundant features (Schiezaro & Pednini, 2013). These techniques are applied to all types of datasets to predict the main features and are an essential step used in classification, cluster analysis, and image retrieval.
Besides, data in healthcare industries is growing exponentially. Much research has been carried out in many directions such as context-aware remote e-healthcare (Sharma & Kumar, 2019a), service level agreement and energy cooperative cyber physical system (Sharma & Kumar, 2019b, Sharma et al, 2019, Sharma & Kumar 2019c), and multimedia healthcare data processing (Rathee et al, 2019). Similarly much work has been carried out in feature selection and disease diagnosis. Many algorithms about feature selection have been proposed (Das & Liu, 1997; Jain & Zongker, 1997; Kohavi & John, 1997; Guyon & Elisseeff, 2003). Recently, evolutionary and swarm intelligence algorithms are employed to optimize feature selection combination is very important. With an optimal feature subset, it is possible to gain good prediction accuracies with low computational complexity (Palanisamy & Kanmani, 2012). A survey on six biologically inspired swarm algorithms, namely particle swarm optimization (PSO), ant colony optimization (ACO), artificial fish swarm algorithms (AFSA), artificial bee colony algorithms (ABC), firefly algorithms (FA) and bat algorithm (BA) and its application in feature selection is also carried out (bin & binti, 2014). Schiezaro & Pednini (2013) model the ABC algorithm for feature selection and classified different data sets. The model achieved a higher classification accuracy of 87.10% on reducing the number of features for hepatitis dataset and 98.26% for labour dataset (Forsati, Moayedikia & Keikha, 2012). It is experimented over a benchmark data sets and demonstrate that applying ABC in feature selection is a feasible method. Palanisamy & Kanmani (2012) have discussed ABC approach for optimizing feature selection. The model reduces feature size and improves in classification accuracy with low computational complexity (Shokouhifar & Sabet, 2010; Uzer, Yilmaz & Inan, 2014).
Further, two major conflicting objectives of artificial bee colony optimization on feature selection is also discussed in the literature (Hancer et al, 2018). In 2019, an improved artificial bee colony technique based on elite strategy and dimension learning is introduced (Xiao et al, 2019). The improved binary artificial bee colony algorithm is analyzed over microarray data for feature selection (Wang & Dong, 2019). Similarly, an integrated model integrating artificial bee colony and gradient boosting decision tree is carried out for feature selection (Rao et.al, 2019). Furthermore with new versions of onlooker bee, natural selection methods for artificial bee colony is introduced for feature selection (Awadallah et al, 2019). But, all these feature selection methods fails to handle uncertainties. Simultaneously, rule generation while analysing real life problems is not addressed.