Enabling Supervised Machine Learning Through Data Pooling: A Case Study with Small and Medium-Sized Enterprises in the Service Industry

Abstract

Insufficient amounts of historical data present a major challenge in real world supervised machine learning projects. Small and Medium-Sized Enterprises (SMEs) are particularly handicapped regarding the collection of historical data. A possible solution to this problem is data pooling, where data from different entities is combined to create larger datasets that are more suitable for supervised machine learning. In this study, we investigate the potential that data pooling has for six companies from the service industry located in Germany and Austria. We find that in the studied scenario each company can benefit from the other companies’ data under certain circumstances. In addition, while most companies benefit from a model that is trained with the data of all other companies, this is not always the case. This is because of specific business characteristics that can significantly affect datasets. In such a case, the key challenge is to determine which companies’ data to include in the pool, i.e., to define the pooling strategy. Therefore, we analyze all possible pooling strategies in our scenario and explain selected results with insights from data distribution and feature importance analysis. We conclude that the consideration of business and data characteristics is critical to the selection of an appropriate strategy.