Lists
Lists
Shrestha, Subeksha (2023) Machine learning algorithms for identity resolution to detect fake IDs in crime data. Doctoral thesis, London Metropolitan University.
This research focuses on applications of various machine learning techniques on an anonymized policing dataset used in the EU SPIRIT Horizon 2020 project to detect fraudulent identities and help Law Enforcement Agencies in the investigation on crime to resolve identity issues. Lack of quality and volumed data, limited methodologies to carry out research on criminal data are some common reasons for fewer research in identity resolution. Crime data are very sensitive data to work and minor inaccuracy in predictions cause massive impact when the system is deployed in real-time as genuine people could be questioned whereas criminals could be sent free. Both issues are addressed in this research, with application of 5 different machine learning models which includes Multilayer Perceptron with TensorFlow and Keras, Support Vector Machine, Naïve Bayes, K-nearest Neighbours and Long Short-Term Memory. The research also focuses on having a comparative study to evaluate these 5 models and select the 2 best models to cascade and improve accuracy in prediction of fraudulent identities.
The main objective of the research is to predict the 5 main suspects who have manipulated their records and are the fraudulent identities out of 39 million records in the policing dataset. To compare the results and test accuracy of the models implemented, various fine tuning of parameters is initiated along with application of suitable optimization and activation functions. Additionally, measures are adopted to avoid overfitting and develop the cascaded model to have higher accuracy to understand criminal records and gain more insights on the different crimes committed, common. age groups that are targeted, gender and role of people associated with the crimes which are the crucial factors associated with false identities that help criminals disguise and manipulate their details easily.
 |
View Item |