Identifying Pitfalls in Machine Learning Implementation Projects : A Case Study of Four Technology-Intensive Organizations

Abstract: This thesis undertook the investigation of finding often occurring mistakes and problems that organizations face when conducting machine learning implementation projects. Machine learning is a technology with the strength of providing insights from large amounts of data. This business value generating technology has been defined to be in a stage of inflated expectations which potentially will cause organizations problems when doing implementation projects without previous knowledge. By a literature review and hypothesis formation followed by interviews with a sample group of companies, three conclusions are drawn from the results. First, indications show there is a correlation between an overestimation of the opportunities of machine learning and how much experience an organization has within the area. Second, it is concluded that data related pitfalls, such as not having enough data, low quality of the data, or biased data, are the most severe. Last, it is shown that realizing the value of long-term solutions regarding machine learning projects is difficult, although the ability increases with experience.