Irina Kolesnikova
March 31st, 2022
The Estonian Police and Border Guard Board (PBGB) reacts to up to 140,000 events a year. About 55,000 of these are events demanding immediate police involvement, where every minute needed to get there is crucial. It is imperative to schedule police patrols placement to be as close to the tensest locations as possible.
The current system of placing police patrol is too time-consuming and does not factor in all the new information. In addition, it is a multiple-stage process, and it takes longer-term human resource planning and shorter-term actual location decisions. All these circumstances lead to suboptimal patrol placement, which has to be improved.
A proper database and AI model are needed to forecast the emergency calls based on a number of factors, starting from historical data of issues and emergency calls and ending with weather conditions at the location.
At the moment, there are two high-priority demands to enhance the inside processes of PBGB work with AI:
In 2022, PBGB has partnered with MindTitan to implement two projects and, thus, provide better services and make Estonia a safer place.
The project is funded by the European Regional Development Fund (85%) and national co-fund (15%).
Approximately 350,000 cases are entered into the police procedure system each year, where in addition to the classifications the data is also described in free text in the form of explanations and solutions. In addition, some data is sensitive (such as addresses and names), which means that the use of this data within the PBGB at the moment is limited by data protection rules.
Thus, the possibility of anonymizing personal data in the free text of case descriptions will ensure the protection of human privacy and personal data. If anonymized and classified in the best way possible, this powerful source of information could help prevent crimes and make police services more effective.
Problem: A lot of valuable data is not used
The solution: the free text analysis model
Second, a task-specific name entity recognition (NER) model will sift through the data, replacing sensitive information with neutral placeholders. Personal names and addresses found will be uniquely anonymized for each text (case).
Thus, the PBGB can use the cleaned data to improve services.
On average, during a 24-hour period, patrols of PBGB come to the aid of people up to 400 times. To reduce the time to get to someone in need, patrols should be in the right places.
Today, patrols are fielded in two stages. First, human resources are planned at the strategic level significantly earlier (e.g., monthly work schedules and 12h shifts). Right before the patrol changeover, a work area is decided at the operational level (e.g., with an accuracy of 1-2 hours).
When planning patrols’ placement, the most important first-stage factors are the time-related ones: day or night, weekend or working day, which season?
At the second stage, historical data about the number of issues in similar periods is processed. Even though patrol schedules are now based on the most crucial historical data, there is no current possibility to take into account other changing factors, such as people’s migration or economic activity.
The problem: How to be in the right place at the right time
The solution: the emergency call forecast model
Forecasting is an AI technique that takes data and predicts possible future values of the data by looking at its unique trends. In our case, algorithms forecast the locations where the patrol is needed the most, based on case histories from up to four years prior, also taking into account economic situation, demographics, and even weather conditions.
During the project, the forecasting model will be created and the obtained result will be available for the e-police application Apollo and the relational part of the PBGB analysis module via a web-based application programming interface.
With the help of the e-police device (Apollo), a patrol police officer working in the field will see the result of the forecast. The model will be able to show both short-term (for example, hourly) and longer-term (for example, one week) forecasts.
Previously generated longer-term forecasts will be refreshed with each update. To present the work process of the model and the intermediate results to PBGB, a visualization tool will display forecasts.
With the free text analysis module, a better-classified database will increase PBGB’s ability to use more data in future innovation projects and research.
The AI-powered tool will provide an operational, multifactorial, and dynamically changing forecast of emergencies that will allow PBGB to meet the ever-growing challenges in today’s world.
Read more about other AI projects that the Estonian government is launching.
