Feature Analysis Using Machine Learning – An Application on Smartphones

Abstract

Due to increased competition in today’s market, catering to diverse customer needs and preferences became a top priority for successful firms. A critical task in the early stages of product development (PD) is determining the configuration of product features. If the features offered are not aligned with consumers’ tastes, the product will likely fail in the market. Therefore, in this thesis, we propose an integrated framework that makes use of publicly available data in order to analyze the features of a product from a customer’s point of view. Given a set of values of the features of a certain product, along with the price and the number of hits, we train various supervised learning models (namely, regression models and decision trees) to predict the two dependent variables: price and number of hits. Depending on the data available, the number of hits can be linked to the number of likes, clicks, or even number sold. This price-based approach, rather than cost-based, enables developers to determine the features mostly affecting the price and the number of hits for a particular product. Using the results of the supervised learning models, two use cases are demonstrated in the context of PD. Firstly, a recommender system is formulated in the form of a mathematical model to be used as a tool by both customers and manufacturers. The main objective of this recommender system is not only to determine the closest product based on the given features, but also to record, and ultimately minimize, the discrepancies between the features offered and the features needed by the customers. Second, a direct value method (DVM) survey is sent out to consumers to compare the feature values from the customer’s point of view to that of the machine learning prediction models, which represent the manufacturer’s point of view. Finally, our methodology is demonstrated using a web-scraped dataset on smartphones, as a case study. The results of the case study confirm the usefulness of machine learning in analyzing product features.