How Artificial Intelligence Helps Crime Prevention
According to a study released by FBI, there is an annual increase of 4.1% in violent crimes and 7.4% in motor-vehicle thefts in the United States in 2016. Despite having stringent laws and intense monitoring, the crime rates seem to be only increasing. Therefore, the best approach to prevent crime is a proactive approach rather than post-crime investigation and action. One of the powerful tools that can help the security department with prevention of crimes is Artificial Intelligence (AI) enabled “crime prediction” methodology. It not only saves millions of citizens from violence and crime but also aids in the better utilization of the limited law enforcement resources.
What makes crimes predictable?
As per a study conducted by the University of California, it is observed that crime in any area follows the same pattern as that of earthquake aftershocks. It is difficult to predict an earthquake but once it happens, the aftershocks that follow are quite easy to predict using patterns and past data. Same can be applied for crimes happening in a particular geography.
Experts believe that a criminal tends to use a method, time and location that has proven successful to them over time. They tend to be in their comfort zone and operate under similar conditions again and again due to their prior experience and minimize the risk involved. This makes them predictable.
As per the Chief of the Los Angeles Police Department, “The predictive vision moves law enforcement from focusing on what happened to focus on what will happen and how to effectively deploy resources in front of crime, thereby changing outcomes.”
Rule-based engines vs. Machine Learning
With the traditional rule-based engine solutions, rules have to be updated frequently which increases manual intervention and are error-prone. Also, with increasing data, rule engine becomes heavy and maintenance becomes tedious.
Machine Learning (ML) builds intelligence from various sources of data. ML algorithms then try to mimic human intelligence and draw patterns and behaviours from the data without any manual intervention. This intelligence gets upgraded over time as new or additional data is being generated from the sources. As new data is added to the system, the ML algorithm automatically adjusts the parameters to check for possible changes in patterns.
Applying analytics to crime data
In simple terms, crime prediction using analytics and machine learning involves integrating data from disparate sources to analyze them and find patterns and behaviors that are repetitive in nature. This enables the police department to draw conclusions about the crimes committed by seasoned criminals, in various locations, and during different periods of time. A huge amount of data is being used for analysis such as historical data, data from CCTV, social media conversations, weather reports, population data, public events data, economic growth-related data, etc. This data is then analyzed through the right set of mathematical models, predictive analytics technique and machine learning algorithms to identify patterns of crime that otherwise can’t be obtained.
The collected data is pre-processed and analyzed to identify the hidden patterns and derive correlation between crime type and locations. Predictive models are built using machine learning algorithms to predict the future crime occurrences.
Techniques used in AI-enabled crime prediction
It is crucial to identify the type, location and method of crime in order to prevent it. The below matrix would help the security experts to choose the right ML algorithm for the required function. For e.g., Random Forrest is used to analyze/predict the “when” and “where” of the crime. To predict the next possible crime scene, the hot spot analysis would be the most suitable choice.
Analytics and ML in crime prediction
Below is a scenario that depicts how an agency can predict crime in advance and alter the outcome. A US city leveraged the benefits of Analytics and ML to reduce burglaries by 30%.
Benefits of using artificial intelligence and machine learning in crime prediction
Accuracy of 60-74% can be achieved in predicting category of crime by multi-label classification techniques like Gradient Boosting Machine and Random Forest.
Crime prediction accuracy of 65-72% can be achieved by analyzing just 4-5 years of crime data.
Including feeds coming from social media, the accuracy of prediction can be increased by up to 15%
By Sarvagya Nayak
Sarvagya is an experienced business manager at Prodapt’s Telebots RPA division with a demonstrated history of building & delivering actionable insights on Analytics, Robotic Process Automation, O/BSS and IoT. His areas of interest are analytics, process improvement, and business model innovation.
