Article Preview
Top1. Introduction
In the modern world, tremendous data are generated in every business. Having a lot of data related to a concerning decision making and problem solving domain is a highly appealing situation. However, the data are generated in a collaborative form via multiple means, multiple approaches, and heterogeneous formats. Obviously, these data come in large volume too. To effectively utilize the data, there is a need for an intelligent process. The application of Artificial Intelligence (AI) can offer some help in this direction. However, traditional artificial intelligence methods such as expert systems can not be of great help because of many reasons. As conventional artificial intelligence based systems are based on knowledge, they are hard to characterize, voluminous, and dynamic in nature.
Further, when the domain is dynamic and undergoes frequent adaption, stored knowledge is quickly absolute. In this case, modern machine learning techniques might be useful. Some examples of such modern machine learning techniques are artificial neural networks, fuzzy logic and genetic algorithms. Such methods overcome the limitations of the conventional artificial intelligence based systems. Each modern machine learning methods has its pros and cons. Fuzzy logic is good at uncertainty handling and approximate reasoning. The artificial neural network approach enables the storage of knowledge in implicit ways and offers advantages of self learning from a large amount of data. Hybrid intelligent methods based on modern machine learning techniques offer advantages of integrated techniques and overcome limitations associated with a standalone machine learning method. This paper discusses the use of hybrid neuro-fuzzy systems in the selection of fast moving consumer goods.
Humans generally and subconsciously categorize and classify things into classes without specific boundaries. For this, linguistic words are being used often. Some examples of these are expensive products, rich people, and luxury cars. Generally, a non-algorithmic way is used to select entities such as fast moving consumer products, documents, jobs, matrimonial profiles, mobiles, garments, finding beneficiaries of loan, Government schemes, etc. Consider a scenario where a fast moving consumer product such as a shampoo bottle needs to be purchased from a superstore. Varieties of products are available in this category with several schemes, plenty of subcategories, multiple brands, and different prices. Similar products are available online too, maybe with handsome discounts. ‘Buy one get one free’ offer on a shampoo bottle may have a special effect on the consumer. However, if the product is close to its expiry date, those two shampoo bottles bought under the scheme may not be used fully. It is very tedious to read all the labels, compare the product with other products in the same store and other marketplaces to make an effective decision as it involves multiple criteria. Above this, such criteria vary for each product. Such scenarios are classic examples of multi-criteria decisions from large and heterogeneous data sources.
Instead of applying the traditional models for the multi criteria based decision making, it is proposed to use the hybrid intelligent technology to offer more human-like decision making and approximate reasoning. To handle vagueness, uncertainty and provide approximate reasoning and inference, fuzzy logic is proposed. Some criteria cannot be mapped from fuzzy to crisp but require fuzzy to fuzzy mapping. In this situation, the use of type 2 fuzzy logic is proposed. To learn the trend of decision making and to know about consumer purchase patterns, the use of an artificial neural network is proposed. Both fuzzy logic and the artificial neural network are supposed to work in a cooperative manner. The hybridization, membership functions, and other technical details are presented in detail in this paper’s subsequent sections.
The paper is organized as follows. Section 2 of the paper discusses the literature survey presenting state of the art in the domain to make such recommendations using various intelligent approaches. The section also enlists observations and common limitations to identify the research gap. Section 3 presents a generic approach for effective and intelligent decision making by using a hybrid type 2 neuro-fuzzy architecture. Section 4 discusses an experiment based on the proposed architecture to recommend fast moving consumer goods, considering multiple criteria to analyze the results. Section 5 presents the conclusion by giving benefits, applications of the proposed approach in other areas with research possibilities ahead.