Under EDA analysis, Hypothesis testing as well as Association Rule Mining, there are some significant relationships: Days and crime types, Education level and crime types, Obesity rate and crime types, Smoking rate and crime types, New moon and crime types, Rainy Weather and crime types. However, for rainy weather, education levels, new moon, these two factors are not significantly correlated with all crime types. For example, for New moon, it is only correlated with Assault and Robbery crime types.
In order to better investigate the potential relationship among different crime types. A network analysis is conducted, which considers the top two crimes happened in one specific day for each zip code. The results show that there seem to exist two clusters for the 8 crime types. Specifically, the crimes can be clustered into: Assault and Drug, which have relatively lower degree; and the rest of the crime types, which have relatively higher degree. This implies that the chance for Assault and Drug crime to be the two most frequent crime is lower than than other combinations, which could also indicate that Assault and Drug will associate with other crime types more often individually. And from the network graph, Theft and Burglary also have a strong correlation between each other, which are connected by thicker edges. The thick edges connecting the two crime means that weight of these network is much heavier and the possibility for the two crime to happen together is much higher than other crime combinations. This is because Theft and Burglary are the two most common types of crime seen in daily life, require similar conditions for commiting the crime, and they share similar traits and have similar patterns during the week. Both tend to happen less frequently during the weekends, and more during weekdays.
Additionally, spatial temporal pattern results are crucial in the context of situational crime prevention. This project applied the monte carlo simulation method to investigate spatial temporal pattern for different types in days in the week. The results show that there are distinctive temporal and spatial patterns for different days of the week. As such, for the predictive modeling, it was based on zip code units and considered day and month as features in order to spatially predict different types of crime in day.
In this case, the predictive analysis modeling are with the features: Day, Month, Moon phase, Moon illumination, Education level, Rainy Weather, Max Temperature, and the predictive modeling in this project was performed for different types of crime to predict the frequency level of specific crime type (If this specific crime type would occur as the most frequent crime type in this space time pattern or not). After model selection and evaluation, the predictive models for all different crime types are with more than 70% accuracy. In this case, these features tend to predict well for future prediction in specific crime types. To know if the specific crime type would be highly frequent events in specific day and particular location would facilitate crime department to make decision for crime prevention. Furthermore, for people who wants to travel to some particular places, this comprehensive prediction modeling is able to help them to know which type it would most likely to occur in these days.
In a word, for crime, which is related to a variety of factors, and also it varies in different crime types, time and location. Therefore, it is really important to take space-temporal and different crime types into consideration. Consequently, accurate prediction based on location, time and crime type, as the predictive modeling in this project, is really critical for the success of any crime prevention initiative.