Abstract. Machine learning models are ubiquitous and impacts everyone. Online shops, for example, have features such as “customers who buy A, also buy B” to boost customer experience and sales revenues by suggesting (hopefully) matching products. While machine learning is crucial for many companies, machine learning models are often treated asmagic, always correct black boxes not looked at by testers and quality assurance professionals. This poses a high risk. Thus, we elaborate in this article thecore conceptsthat enable testers to discuss with data scientists about quality assurance in machine learning.
1 Understanding the Context
The idea behind machine learning algorithms is that applications predict outcomes –without being explicitly programmed so –after training them with historic data. [Bur20] Typical examples are:
- Image classification: Does the picture contain a dog or a cat?
- Object detection: What objects are on the picture and where are they?
- Predicting sales opportunities: Whom should we try to sell a pink flamingo?
- Sentiment analysis: Are the reviews positive or negative?
Machine learning models are rarely used stand-alone. They are often part of a larger software system, also referred to as embedded analytics.
The complete article appeared in the magazine The Tester of the BCS – The Chartered Instsitute. You can read and download the article Quality Assurance and Machine Learning Models as Pdf here or from their webpage.