A hybrid machine learning process framework for data-driven direct product marketing

Abstract

The ability to propose the right product, to the right customer, at the right time is valuable in marketing. Direct marketing strategy entails the analysis of customer patterns to predict the best product offerings that are relevant to a customer per time. Machine learning (ML) has proven to be an improved, more effective approach for direct marketing compared to using more traditional statistical methodologies. However, the disadvantage of ML models is that they normally perform in a black box manner, where the predictions are produced without an explanation. In many cases an expert is required to interpret the outputted result. Studies have shown that this lack of understanding and explanation leads to reduction in the user’s confidence in the results produced. This situation has resulted in a reluctance of the managements of different companies to implement ML-based solutions. The need for highly skilled personnel to interpret outputted results also increases a company’s operational and capital expenditure. Thus, the need for ML-based solutions for direct marketing that produces accurate results that are accompanied with relevant explanations, is critical. The aim of this study was to develop a hybrid machine learning process framework that could facilitate data-driven direct product marketing. The hybrid ML framework leveraged the statistical processing strength of ML models and addressed the lack of explanation by adding intelligent reasoning. The intelligent reasoning component was used to generate relevant explanations that accompanied the predicted outputs that are generated by a selected ML model. The hybrid framework has the capability to utilize multiple ML models. The applicability of the hybrid framework was demonstrated by using an anonymized and de-identified dataset from a South African-based telecommunication company. The study adopted a design science, research strategy as the main methodological approach for the formulation of the research design. To achieve this, an analysis of existing literature, and interactive sessions with domain experts were conducted to gain an accurate understanding of requirements. This then formed the basis for evolving the architecture and the design of the hybrid ML framework. The implementation and evaluation of the prototype system was done within the framework of Design Science Research (DSR). The hybridized ML framework consists of three ML models, which are Support Vector Machine (SVM), Random Forest (RF) and Artificial Neural Network (ANN). The intelligent reasoning component of the hybrid framework is composed of Case-Based Reasoning (CBR) and Rule- Based Reasoning (RBR). It also has a user-friendly interface that allows a user to select a specific operation/function that is desired, through which results of operations and explanations are also presented to the user. The hybrid ML framework was evaluated using the Goal Question Metric (GQM) approach in order to determine its performance and usability. The evaluation results obtained from feedback of participants show that the results are adjudged to be very accurate, while the quality of explanations is also satisfactory. The study reveals that data-driven, direct marketing through the application of explainable AI methods is possible and is valuable to the telecommunication industry.