Abstract:
Crime is an intentional and unintentional act that causes physical harm, mental harm, property
damage, or loss. Personal crimes are the type of crime against a person; it is a crime that is
directed at an individual and harm or injury can be traced to the victim, including murder,
aggravated assault, rape, robbery, and others. This study attempts to develop a personal crime
prediction model by using a machine learning approach. The proposed predictive model is
trained on a dataset that includes personal crime cases that occurred from September 2010 to
April 2015 E.C in Jimma Zone. The dataset was collected from secondary sources of data from
Jimma Zone high court, which includes 6000 instances, and 16 attributes of the cases were used.
That data contains detailed information about criminals and types of personal crimes. The
algorithms used in this study are machine-learning algorithms DT, RF, NB, and RF. The dataset
was split into training and testing 80%:20% datasets respectively. The decision trees have the
highest evaluation metric precision of 85 values, the random forest has a recall of 98, naive
Bayesian has a recall of 88, and Extreme Gradient Boosting (XGB) has a recall of 79 values.
And the researcher also took another training and testing dataset 85%:15% respectively, and the
decision tree has the highest evaluation metric with a recall of 85 values, the random forest has
a recall of 98 values, naive Bayesian has a recall of 58 values, and Extreme Gradient Boosting
(XGB) has the highest evaluation metric is a recall of 98 values. From these models, random
forest and Extreme Gradient Boosting have almost the same evaluation metrics. But, the random
forest has the highest evaluation metrics on the two training and testing datasets. Based on this,
the model was deployed using a random forest model. Generally, this study tried to achieve the
personal crime predictive model in spatiotemporal by using a machine learning approach.
