Validating Business Problem Hypotheses: a Goal-oriented and Machine Learning-based Approach
Validating a business problem hindering a business goal is often more important than finding solutions to the problem, specifically during requirements engineering. For example, validating the impact of a client’s low account, high transactions, or high loan payment per month for a client’s unpaid loan decreasing the loan revenue of one bank would be critical as an information system can be designed for the bank to take some actions to mitigate the loan default. However, many business organizations are struggling to confirm whether some potential problems hidden in Big Data are against a business goal or not. In other words, they face difficulties finding real business problems and then improving business value, although the investment in Big Data and Machine Learning (ML) projects has increased. One challenge might include a lack of understanding about relationships between business problems and data. The other challenges might consist of determining a testable factor associated with a potential business problem, preparing a relevant dataset corresponding to the business problem, analyzing the impact on the business problem to other problems, and reasoning about inter-connected problems. Information systems solving unconfirmed problems frequently tackle an erroneous problem and give incorrect predictions, leading to some dissatisfying systems, consequently not achieving business goals, even redeveloping the systems and taking many business resources. This dissertation presents a Goal-Oriented and M achine learning-based framework using the notion of a Problem HY pothesis, Gomphy, to help validate potential business problems. We propose five main technical contributions: 1. The domain-independent Gomphy ontology and process are presented explicitly and formally for describing categories of essential concepts and relationships concerning business goals, problems, problem hypotheses, ML, and a dataset. The ontology ensures that business goals and the related business problem hypotheses are traceable to an ML dataset. 2. An entity modeling method of a problem hypothesis is elaborated to help capture business events and determine testable factors. 3. A data preparation method is described to build an ML dataset, mapping a concept of a problem hypothesis to a data feature. 4. A feature evaluation method is presented using ML and ML Explainability library to detect contribution relationships among the business hypotheses and problems. 5. A set of formalized validation rules are described for reasoning about connected problem hypothesis validation in a goal-oriented problem hypothesis model. To see the strength and weaknesses of the Gomphy framework, we have validated potential banking problems about an unpaid loan and customer churn in one retail bank as empirical studies. We feel that at least the proposed framework helps validate business events that negatively contribute to a goal, providing insights about the validated problem.