University Links: Home Page | Site Map
Covenant University Repository

AN INDEX MAPPING-BASED DEEP TRANSFER LEARNING APPROACH FOR VIOLENT CRIME PREDICTION

Falade, Adesola and Covenant University, Theses (2020) AN INDEX MAPPING-BASED DEEP TRANSFER LEARNING APPROACH FOR VIOLENT CRIME PREDICTION. ["eprint_fieldopt_thesis_type_phd" not defined] thesis, COVENANT UNIVERSITY.

[img] PDF
Download (438kB)
[img] PDF
Download (438kB)

Abstract

Crime has been with us from time immemorial and impacts negatively on the quality of life of citenzery and the general health of a nation. Different approaches have been used in the previous studies for violent crime prediction to aid predictive policing, making conventional policing more efficient and proactive. The violent crime rate in Nigeria has been on the continual increase. There are dearths of state-of-the-art measures like the predictive policing approach for violent crime prediction in Nigeria. The existing controlling and preventive measures for tracking and controlling violent crime are not sufficient. The violent crime predictive models in the previous studies do not have sufficient features to predict violent crime types and time-slot of violent crime occurrences. More so, violent crime occurrence prediction is nearly impossible when there is insufficient dataset. Therefore the aim of this study is to develop an improved predictive model for violent crime prediction using an index mapping-based deep transfer learning approach with a view to increasing the predictive accuracy of violent crime prediction in Nigeria. The sources of data for this study are from historical violent crime records of Nigerian Police, and online reported violent crime data. As a prelude to generating training data, cleaning of data and relevant feature selection of violent crime dataset were performed. Consequently the predictive model developed during the empirical study was trained on IBM Watson Machine Learning studio. The model consists of four layers: data collection, features extraction, spatio-temporal violent crime prediction premised on index mapping-based deep transfer learning, and violent crime hot spot visualizer layer. The evaluation experiment was conducted through benchmarking with the existing approaches using accuracy, semantic precision, and recall as well as F-measure metrics with the use of confusion matrix. The violent crime prediction model evolved in this study delivers a predictive accuracy of 92.53% across the six violent crime dataset used. This result showed that an index mapping-based deep transfer learning model developed outperformed other Machine Learning models used in the previous studies. In addition, the proof-of-concept web-based application for reporting and alerting violent crime developed was also evaluated through a system usability scale survey with a result of 85.28%, which represent usable system with good usability rating score. The study therefore serrves to benefit the citizens of Nigeria by alerting them of violent crime hotspot areas in the country, and also would enable police authority develop violent crime prevention strategies that could mitigate spate of criminal activities in the country.

Item Type: Thesis (["eprint_fieldopt_thesis_type_phd" not defined])
Subjects: Q Science > Q Science (General)
Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Divisions: Faculty of Engineering, Science and Mathematics > School of Civil Engineering and the Environment
Depositing User: Mrs Hannah Akinwumi
Date Deposited: 01 Mar 2021 21:04
Last Modified: 01 Mar 2021 21:04
URI: http://eprints.covenantuniversity.edu.ng/id/eprint/13858

Actions (login required)

View Item View Item