Predicting Customer Behavior with Combination of Structured and Unstructured Data

Abstract

Presently, there are numerous e-marketing and m-marketing mediums that exist such as YouTube, SMS, What Sapp, Google, twitter, yahoo, Facebook, LinkedIn, email and personal blogs. These mediums are beginning to be used for marketing purposes, particularly by the SMEs in Nigeria. The aim of this research is to address the problem of deciding which of the mediums mentioned above is mostly appropriate to target customer of a particular SME and also to discover the type of data that is most appropriate for analysis in making this decision. In order to achieve this, data was gathered by administering questionnaires and pre-processed based on structured and unstructured data sources. The J48 decision tree classification algorithm was used to mine the data, relevant predictions were made from the structured and unstructured data and the results were evaluated. The results revealed that predicting from unstructured data expresses more of popular opinion, so decision can start from unstructured results and be fined tuned or validated with predicting from structured data. Though structured prediction appears to be better than unstructured, unstructured prediction is still very valuable in situations where there are no structured data such as analysing text messages. Also, Models developed for predicting customer behaviour as regards the marketing channels studied, will form the foundation for marketing decision making, in small and medium businesses in Nigeria.