Discrimination-aware data analysis for criminal intelligence

The growing use of Machine Learning (ML) algorithms in many application domains such as healthcare, business, education and criminal justice has evolved great promises as well challenges. ML pledges in proficiently analysing a large amount of data quickly and effectively by identifying patterns and providing insight into the data, which otherwise would have been impossible for a human to execute in this scale.
However, the use of ML algorithms, in sensitive domains such as the Criminal Intelligence Analysis (CIA) system, demands extremely careful deployment. Data has an important impact in ML process. To understand the ethical and privacy issues related to data and ML, the VALCRI (Visual Analytics for sense-making in the CRiminal Intelligence analysis) system was used . VALCRI is a CIA system that integrated machine-learning techniques to improve the effectiveness of crime data analysis. At the most basic level, from our research, it was found that lack of harmonised interpretation of different privacy principles, trade-offs between competing ethical principles, and algorithmic opacity as concerning ethical and privacy issues among others.
This research aims to alleviate these issues by investigating awareness of ethical and privacy issues related to data and ML.
Document analysis and interviews were conducted to examine the way different privacy principles were understood in selected EU countries. The study takes a qualitative and quantitative research approach and is guided by various methods of analysis including interviews, observation, case study, experiment and legal document analysis.
The findings of this research indicate that a lack of ethical awareness on data has an impact on ML outcome. Also, due to the opaque nature of the ML system, it is difficult to scrutinize and as a consequence, it leads to a lack of clarity in terms of how certain decisions were made. This thesis provides some novel solutions that can be used to tackle these issues.