Browsing by Author "VERONICA TESFAYE BELAYNEH"

Filter results by typing the first few letters
Now showing 1 - 1 of 1
  • No Thumbnail Available
    Item
    PREDICTION AND ANALYSIS OF CRIME AGAINST WOMEN USING MACHINE-LEARNING
    (Hawassa University, 2024-07-12) VERONICA TESFAYE BELAYNEH
    Crime is a widespread issue globally and a serious concern affecting the lives of women. This study aims to employ ensemble learning methods to predict crimes against women in different areas of Hawassa City. The proposed research is based on analyzing crime patterns in previously referenced years. By combining predictions from multiple models, ensemble learning aims to improve the overall performance of the model. The results of the study indicate that the ensemble of machine learning models, particularly the proposed classification models, significantly improves crime prediction compared to traditional methods, as evidenced by improved performance metrics. The study uses a dataset of 3,318 records, with 454 entries from a One Stop Center, and considers 8 attributes: 'Subcity', 'Kebele', 'Year', 'Age', 'Marital status', 'Time of crime', 'Categories', and 'Crime'. The focus is on predicting specific crime types: forced use, kidnapping, snatching, beating, insult, rape, theft, and threats. The developed model i s based on an ensemble technique which is a Machine Learning (ML) based approach. The model has good accuracy, which is 92.87% accuracy. The accuracy results of various machine learning classifiers for predicting crimes against women in Hawassa City are as follows: Random Forest achieved the highest accuracy at 92.87%, followed by Logistic Regression at 91.57%. The Decision Tree classifier also performed well with an accuracy of 91.37%. Both the SVM and Voting classifiers attained an accuracy of 90.86%. AdaBoost had the lowest accuracy among the evaluated models, with a score of 89.86% In addition to the accuracy, based on the importance of each input feature to the final model, Crime categories take the main influencing share by 70.6% from the total input parameters. According to the gathered data, the model demonstrates notable precision and recall rates when employing the Random Forest classifier, especially in effectively identifying specific classes with high precision and recall. The study contributes to changing traditional police stations and investigating activities by incorporating new technology and data recording methods that are more suitable for contemporary scenarios.