Report on Data Analytics Initiatives in West Midlands Police (WMP)
To: The new Chief of Police, West Midlands Police Force, Police Headquarters, Lloyd
House, Colmore Circus Birmingham B4 6NQ
Introduction (a primer on the county, its boroughs, and WMP):
West Midlands is the second-most populous county in England with an estimated population of 2,939,927 (2020). It is metropolitan and follows a combined authority style of local government. Along with 6 other counties, it forms the official West Midlands region in the country, and though sometimes referred to as "Greater Birmingham", it constitutes 7 urban boroughs or areas - the cities of Birmingham, Coventry, and Wolverhampton, and the boroughs of Dudley, Sandwell, Solihull, and Walsall[1]. The police force in West Midlands is territorial and not federal[2]. The force covers an area of 348 square miles (900 km2) for its ~2 million inhabitants. As of 2020, there are 6,846 officers, 484 police community support officers (PCSO), and 219 volunteer special constables. The force is organized into ten Local Policing Units (LPUs), each of which is served by four main policing teams: Response, Neighbourhood, Investigation, and Community Action & Priority (CAPT), as well as 17 specialty crime teams that include counter-terrorism, traffic, and firearms[2].
Crime in West Midlands:
Historically, crime rate statistics have shown that the West Midlands is the third safest county in England, Wales, and Northern Ireland[3]. Given the size and density of its constituencies, this position is appealing. However, since 2016, the monthly violent crime volume has been increasing and in the last year, our police force witnessed the country's biggest rise in crime amid a surge in violence and sexual offenses. Crime went up by 21% in the region, 2x more than the numbers of the next most hit force in the nation. 137,549 crimes were reported in 2021, out of which violence against people increased by a distressing 45 %, while weapons possession increased by 45 %, stalking and harassing cases doubled, and sexual offenses increased by a third. The number of murders increased by 6% to 53 as well, making it the second-highest in the UK behind London[4]. Some of this is due to the improvements we have made to record incidents, but at the same time, the force has been facing considerable pressure to adequately react to this surge owing to government cuts in funding since 2010[4]. These numbers are alarming but we have also made progress on certain key offenses. Thefts were down 7%, burglaries were down 16%, and criminal damage and arson were down 11%[4]. Concerning population, Birmingham and the Sandwell town of West Bromwich have high crime rates and they continue to be key targets in our enforcement[3].
It is evident that a lot of work is required to proactively reduce these numbers. One essential factor that has consistently helped us in managing and intervening incidents better is data and analytics. The Covid-19 pandemic outbreak in 2020 presented a very different crime landscape and we had to pivot our usual decision-making to safeguard the community while also managing the risk to our own officers in the force. This was only made possible with objective evidence and intelligence. Similar to other police forces in the UK, there was an acceleration in the development of dedicated analytics applications, to support crime forecasting and geo-mapping offenses during the pandemic's intense moments[5]. Within the initial days of lockdown, analytics dashboards were created to help us manage our response. We were able to monitor clusters of high virus exposure, the location of our enforcement activity, develop breach reports as well as observe fixed penalty notices. The last was important since lockdown rules were changing over time and we had to ensure that the notices were fair and consistent[5].
The efforts around the virus outbreak reaffirmed our belief that there is always an untapped potential for analytics in policing. And this idea of intelligence-led policing isn't new, many police departments across the world already use tried-and-true approaches like hotspot policing. With predictive modeling techniques, however, police forces like ours:
Can find patterns in data that point to areas for further investigation.
Explore security threats by studying people, events, and locations.
Deploy personnel to where they are needed the most and control costs, among many other applications.
We were fortunate to understand the true value of predictive policing early on and the crux of this report is to bring to your notice our analytics capabilities and similar investments.
NDAS:
NDAS or the National Data Analytics Solution was established in 2016 with the help of 9 founding partners and 4 police forces that share data (West Midlands, West Yorkshire, Warwickshire, and West Mercia)[6]. NDAS is a new, scalable and flexible analytics platform/system built for UK law enforcement. It uses advanced analytics and A.I (Artificial Intelligence) to deliver insights that can aid in high-priority operational and organizational issues[7]. NDAS is sponsored and mandated by the Home Office but its implementation is being spearheaded here by the West Midlands police force. We entered into a services agreement with Accenture plc [8] to process our data and the system uses a combination of A.I, statistics, and multiple data processing "zones" to assess the risk of people committing crimes as well as the likelihood of someone falling victim to crimes. Police funding has been cut significantly over recent years, and our project managers on this initiative claim NDAS is a system that can immensely help in enforcement - a system that can look at all individuals already known to officers, with the aim of prioritizing those who need interventions most urgently. The idea being individuals flagged by the system to be offered interventions, such as counseling, to avert potential criminal behavior.
We take great pride in NDAS since this is the first such project of its kind in the world. What sets NDAS apart from other predictive policing systems is that it pools multiple data sets from a number of police forces for crime prediction[9]. The team obtained more than a terabyte of data from local and national police databases in the early stages, including records of persons being stopped and searched, as well as crime reports. Around 5 million individuals were identifiable and the NDAS software analyzed the data and discovered roughly 1400 signs that may be used to forecast crime, with approximately 30 of them being very effective[17][6]. These included the number of crimes perpetrated with the assistance of others and the number of crimes committed by others in the same social group as the individual. These signs will be used by NDAS' "Machine Learning" and "Natural Language Processing" or NLP A.I components to identify whether individuals known to the police are on a violent trajectory similar to that seen in previous incidents, but who haven't yet escalated their behavior. A risk score will be finally issued to these individuals, reflecting the possibility of future criminal activity.
NDAS was established to help with 7 use cases and be a stepping stone for more in the future. The 7 scenarios are modern slavery (MS), organized exploitation, firearms, domestic abuse, violent crime, workforce wellbeing, and most serious violence or MSV (people likely to commit their first most serious violence offense in the next 24 months). The development for the last two is currently paused to allow more room for NDAS to be thoroughly tested. Currently, the MS, violent crime, and organized exploitation use cases are fully operationalized. Each use case has a custom, feature-rich, and highly interactive dashboard that can generate quick visualizations and show key insights from data[6].
Successes and Failures:
The modern slavery (MS) use case via NDAS has had a positive impact[6]. Modern slavery is the severe exploitation of people for personal or commercial gain and it is a pressing issue in our society that is institutional and difficult to uncover. For MS, NDAS consolidates data into a POLE model - People, Objects, Locations, and Events. And it optimizes this POLE data model to show a network view of people or "nominals" associated with MS activity[6]. The goal is to discover previously unidentifiable patterns and trends with built-in network analytics and NLP modules. With NDAS, we discovered that on average 1 MS nominal is in fact connected to 54 other nominals. Around 15k intelligence logs were identified as relating to MS and they were previously untagged. The data sources for the MS use case have been consolidated into 3 key data tables taking insight generations for MS analysts from around 200 hours to only a few minutes with the help of NDAS. With MS, the final step was to showcase any insights discovered in a way that's useful and assists the end-user in identifying modern slavery elements. This includes not only network visualizations but also relevant contextual information related to the events and associated people. This entire view is presented at two levels of granularity as well[10]. An operational team can use the low-level granularity to follow up on highlighted situations using current processes. The tool's high-level aggregated granularity lets management teams use it to drive strategic decision-making and gain a better knowledge of modern slavery through patterns and data.
But not all endeavors with NDAS have been successful. The most serious violence (MSV) use case was created to see if someone will commit their first violent crime with a gun or knife in the next two years. People who already had an interaction with the two police forces engaged in the tool's development, West Midlands and West Yorkshire, were assigned risk scores. They are more likely to commit one of the crimes if their score was greater. When MSV was greenlit to be operationalized, it was hit unfortunately with multiple issues. The MSV component fundamentally contained a coding flaw that made it incapable of accurately predicting violence [11]. This coding error was found in the definition of the training dataset which made the entire tool unviable. Before the error was discovered, the system had an accuracy of up to 75% [12] and out of 100 people who were believed to be at risk of committing serious crimes with a gun or knife, 54 from West Midlands and 74 from West Yorkshire were predicted to carry out one of them. Following the error's discovery, this accuracy dropped from 14% to 19% and we had to pause all model operationalization since there wasn't sufficient clarity. NDAS, being a government initiative has been made aware far and wide to media and press organizations. Our transparent policy to keep the public aware of updates has been lauded [9] but there has also been some scrutiny regarding verifying the system's success [12] [13]. These criticisms include NDAS basing predictions on past arrest histories and therefore reinforcing bias in certain places, and whether it was useful for public welfare to intervene when an individual has not even committed a crime just because they are likely to do so in the future [14].
Critical Steps Forward:
1. Address any and all potential ethical concerns (from ATI DEG or otherwise):
As discussed above, there has been plenty of mistrust when predictive analytics has been proposed in law enforcement. There is a devoted debate on the role of bias and biased data and how predictive policing systems can unknowingly target certain groups of people who have been unfortunately historically marginalized in policing [15]. Creating a fair, equitable, and trustworthy system was one of our top priorities when designing NDAS. And to that regard, we invited the Alan Turing Institute Data Ethics Group (ATI DEG) [16] to review NDAS' high-level strategy and help us understand how a suitable ethical framework could be applied to this approach. The team was able to provide us with a complete ethical review of one of our use cases and laid out a governing set of principles we could follow that would allow for more transparency without compromising efficiency in NDAS[17].
We welcome the suggestions and are committed to following a fair and holistic approach to public safety. And while we were able to incorporate many concerns into NDAS [18], there is still a demand for more oversight and intervention in the development of such systems [13]. An immediate critical step would be to address all possible loopholes and ethical concerns (those raised by ATI DEG and otherwise) within NDAS since failure to do so can prove detrimental to NDAS' primary goal to reduce criminality. NDAS has a proven capacity to prevent harm and it needs to be operationalized with more use cases to establish itself as invaluable[6]. As a law enforcement agency that works to promote justice, we believe in the guidance of ethics and hold it with the highest regard. We also believe that any decision regarding ethics should not occur as a result of public outcry or backlash.
With a strong ethical oversight, policing can serve the public in an informed and impartial way. But at the same time, a potential risk could be the blind acceptance of such principles. Policing is not devoid of ethics which the ATI DEG opposes. Such a complex belief can eventually lead to a one-sided functioning in NDAS and allow itself to fall prey to an "influence" the ethical concerns fundamentally raise. While the primary job of the police is to prevent and detect crime, we also spend resources in a variety of areas that are not directly related to crime, such as dealing with people in mental health crises, non-fatal road accidents, homelessness, and suicide [18].
2. Technical and legal data challenges in NDAS:
One of the challenges we faced initially (and continue to do so) in the development of NDAS was the different formats of data among the UK's police forces. Although not insurmountable, this has added an extra level of processing to adapt recorded data extracts into a shared format for NDAS to use. But this problem compounds when supplying NDAS with said recorded data.
The partnering police forces have disparate customs and are reluctant to share data via a secure uniform channel[19]. Because of legislative concerns regarding cloud storage (a computing model that stores data on the internet through a provider who manages and operates data storage as a service[20]) of sensitive police data, obtaining data from different police partners has been a key issue in this project. With multiple stakeholders, there are different legal views on how sensitive policing information should be handled. To that end, we have currently partnered with the Information Commissioner's Office to ensure compliance with data protection regulations
[7][19]. But some continue to prefer handing it over in person on a physical device. In the long term, where a self-sustaining NDAS operation is desired, such processes cannot be considered since the volume of data can exceed terabytes and wouldn't always be feasible. So the critical step here would be to work with our IT partners in designing a communication strategy for uniform data ingestion in NDAS. This can reduce unnecessary overhead and can incorporate a level of automation that is expected from an advanced analytics system at the scale of NDAS.
The obvious risk here however is the valid privacy concerns since the data is essentially sensitive people information. A secure network or channel will be required for communication, one that is robust enough to not be susceptible to cyber-attacks. Another decision that has to be finalized is the choice of delegating data conversion for NDAS or continuing NDAS to include a processing step.
Conclusion
Crime and the continual intent to break the law is an unfortunate element of the day and age we live in. However, this new era has also given us access to advanced technological pathways, the ones used by NDAS - to reach the frontiers of justice faster. While leveraging it is difficult and is bound to bring in criticism, we remain optimistic about data analytics helping us with informed decision-making, better risk assessment, and crime prevention. We look forward to your captaincy and working together with the utmost integrity to serve the great people of West Midlands.