Abstract
An excessive amount of data is daily generated, and the customer’s journey becomes extremely complicated. Industries and decision makers struggle to follow the new trends and they invest huge budgets trying to close the gap between the data and the consumer’s behavior. The need of using artificial intelligence (AI) models which combine marketing data and computer science methods seems imperative. Data mining, machine learning (ML), and deep learning methods act complementary to marketing science through the data classification, the user profiling, the content optimization techniques using data analysis, management, representation methods, and tools to generate highly accurate results. The thesis consists of two parts: the theoretical and the practical. The theoretical part bridges the gap between marketing and informatics engineering by conducting a literature review on cornerstone marketing and computer science definitions including physical and digital marketing, consumer behavior, A ...
An excessive amount of data is daily generated, and the customer’s journey becomes extremely complicated. Industries and decision makers struggle to follow the new trends and they invest huge budgets trying to close the gap between the data and the consumer’s behavior. The need of using artificial intelligence (AI) models which combine marketing data and computer science methods seems imperative. Data mining, machine learning (ML), and deep learning methods act complementary to marketing science through the data classification, the user profiling, the content optimization techniques using data analysis, management, representation methods, and tools to generate highly accurate results. The thesis consists of two parts: the theoretical and the practical. The theoretical part bridges the gap between marketing and informatics engineering by conducting a literature review on cornerstone marketing and computer science definitions including physical and digital marketing, consumer behavior, AI, and ML. It also states the research motivation, scope, significance, and questions. The practical part examines the online and offline consumer’s behavior by using a decision-making method, which analyzes the consumers data, and it helps the decision-makers to understand potential opportunities and needs. A method which combines decision trees (DTs) and genetic algorithms (GA) through a wrapping technique is introduced. This method is known as the GA wrapper and its logic is based on optimal features selection. The used method can generate results through users’ data processing aiming to assist decision makers to take faster decisions. Consumers’ data may originate from the digital or the physical environment including official repositories, surveys, and corporate social media pages insights. The goal is to generate optimized subsets from the initial number of attributes maintaining the initial prediction accuracy. The GA wrapper passes through four stages including design, implementation, verification, and application. Towards the completion of the thesis goal to generate optimal features there are certain objectives which had to be met. Using a recursive approach, the GA wrapper properties broke down into certain objectives. These objectives had to be examined and evaluated prior the GA wrapper model design, implementation, verification, and application phases and they include a series of tests and examinations. The thesis structure is based on recursive reasoning for each one of the objectives. Prior the GA wrapper design stage, there are two case studies of users’ data applied on decision trees classification techniques to show the innovative nature of DTs. Moving backwards before the DTs applications, a comparative analysis between two of the most efficient classification techniques which maximize the advantages of collaborating with a genetic algorithm, the decision trees’ induction and the Bayesian’s learning takes place, to indicate the best technique for the current datasets. Continuing to the earliest stages of this study, a statistical analysis of collected users’ data is conducted to reveal the appropriate data correlations needed for ML classification. These objectives apart from creating a cohesion among the thesis stages they generate new knowledge.Finally, the findings are exceeding the initial author’s expectations. The GA Wrapper manages to generate highly accurate results from a series of data indicating best practices. This research introduces a ML method which can be applied to a wide range of consumers’ or users’ behavior data, helping decision makers perform better strategies, mitigate the decision-making risk, and increase customer’s engagement.
show more
