Algorithmic Bias In Machine Learning Based Marketing Models: The Neural Networks Experience

Abstract

In today’s data-driven landscape of competitive business settings, many digital marketing organizations are grappling with challenges of customer deflection and  poor return on investments. This is due to their inability to address the issues of dynamic customer behaviour, data classification bias and dataset imbalance hence, the motivation for this study. The explosive growth of big data has created both opportunities and leeway for businesses seeking to gain valuable insights from their data assets as to make informed decisions. This paper presents a machine learning based marketing model that utilizes big data analytics and visualization techniques to provide valuable insights for informed decision making in the business sector which facilitates digital marketing process and enhances return on investment. The Wide Area Neural Network technique which is an extension of the traditional neural network with more hidden layers for increased processing capabilities was the methodology adopted for this study. The primary data was collected for 43,750 customers  sourced from MTN Nigeria, then processed using Matlab to generate a secondary data of 20,250  making a total sample size of 64,000 used for the study. These data were used to train the neural network algorithm for the model, - Fittest back-propagation algorithm, which was modeled with genetic algorithm to address the issue of random hyper-parameter selection during the training. The training process was implemented with Matlab software and the model generated was achieved with JavaScript programming language. The result reveals that this model exceled in predicting customer behaviours. This model was applied to recommended  customer packages and performance metrics such as Positive Predictive Value, False Discovery Rate, True Positive Rate, False Negative Rate and Area Under Curve were employed to assess the effectiveness of this model, however, the model stands out, showcasing superior performance in correctly identifying positive cases and minimizing false negatives. The study gives corporate insight  for better decision making, addresses the issue of classification bias and enhances returns on investments.