Auto-Classification of FIA-Cybercrime Wing Complaints Using Bidirectional Encoder Representations from Transformers Model

Abstract

The Cybercrime Wing (CCW) of the Federal Investigation Agency, which was formerly
known as the National Response Center for Cybercrime (NR3C), is governed by rules that were created in
2016 as part of the Prevention of Electronic Crimes Act (PECA) to combat cybercrime. Criminal activities
executed using computers and the internet are referred to as cybercrimes. In order to carry out illegal
activities, cybercriminals make use of any information system as their primary means of communication
with the devices that belong to their victims. This research mainly focusses on the cybercrime complaints
with an automated classification system. To achieve the automatic modelling for classification of different
types of cybercrimes, this study used the Bidirectional Encoder Representations from Transformers (BERT).
Its obstacles mainly include the possibility of human errors as it manually classifies cybercrime complaints,
also that there might be delays during handling in comparison with an automated system. The dataset
includes complaints submitted in English during a two years window, and it was encoded, tokenized and
cleaned thoroughly. The purpose was to simplify the training process for the model. The study used a
lightly fine-tuned, pretrained (BERT)-base-uncased model. The findings confirm that the model can be used
for classifying complaints and exhibits an excellent classification accuracy, precision and F1-scores between
different cybercrime offences indicating its supremacy among advanced Natural language processing (NLP)
techniques to strengthen cybersecurity measures.