Technology

singular perturbationsロゴ singular perturbationsロゴ
Named the Crime (CRIME) Prophet (NABI), this system predicts “when and where future crime will occur”.

Based on past crime patterns and surroundings, you can predict where future crimes are likely to occur. In addition, we will enhance the crime deterrence effect by presenting a route that focuses on guarding areas where crime is likely to occur based on crime prediction.

Improved security effect

Based on the crime prediction results, if you formulate a route that focuses on areas where the probability of crime is predicted to occur, the security effect will be greatly improved. The graph below is the result of calculating and verifying how much the established security route covered the place where the crime actually occurred as a security effect based on the actual data. Verification in Tokyo, Nagoya, and Adachi Ward has shown that it is more than 1.5 times more effective than conventional crime prediction methods. By route formulation, it can be seen that the spatial distribution of the prediction results of CRIME NABI is more effective than the conventional method.

singular perturbationsロゴ singular perturbationsロゴ

Prediction

Crime prediction algorithm

The system collects data on past crimes, artificial density, land use data, weather, etc., and uses two types of proprietary algorithms to predict crimes.

singular perturbationsロゴ
1. Prediction by time information

Criminals have a temporal pattern because once they succeed in committing a crime, they repeat the same technique. By applying the theoretical physics formulation to the model that can describe this, stable calculation is possible even with small data, and high accuracy is achieved.

2. Prediction by spatial information

The model is designed to describe the type of crime to be predicted by adding up various spatial patterns of crime occurrence and population density. The bottleneck of such algorithms is that they are computationally expensive, since they usually require large data input and unknown parameters for the number of spatio-temporal meshes. Singular Perturbations has developed a proprietary mathematical algorithm that pre-compresses the data, enabling fast prediction calculations.

High precision

In compared to conventional methods, CRIME NABI's forecasting method can predict crimes without significant differences from actual crime maps.

singular perturbationsロゴ singular perturbationsロゴ
High speed

As input data increases, CRIME NABI can achieve unparalleled speedups, whereas conventional methods require more computation time. Edge computing is also possible because it utilizes a method that allows the cost function optimization process to be computed in O(N^0). This technology will lead to a future where calculations are performed on mobile devices, drones, and robots to provide optimal security. (free version)

singular perturbationsロゴ singular perturbationsロゴ

Routing

Route formulation algorithm based on crime prediction

Depending on the crime situation, we will develop a route to focus on security where crimes are predicted to occur.
By reversing the crime prediction results, it is possible to formulate a safety route that takes safety into consideration, although it is not the shortest.
This will improve the security effect and reduce damage.
In the figure on the right, you can see that the security routes established in the areas where m crimes are predicted to occur are densely packed.

singular perturbationsロゴ
Main research results Adoption of research projects on crime prediction, dissertations, patents, etc.

Research teams specializing in computational criminology, spatial statistics, computational science, and criminology develop various crime prediction algorithms.

View more
Research consignment /
research grant
2021
  • Tokyo Metropolitan Small and Medium Enterprise Promotion Corporation Next Generation Innovation Creation Project 2020 Grant Program
  • FY2018 commissioned research "Empirical R & D for solving regional issues through data linkage and utilization" (2018.12 ~ 2021.03, evaluation at the end: highest "S")
Papers
"Compositionally-warped additive mixed modeling for a wide variety of non-Gaussian spatial data", Daisuke Murakami, Mami Kajita, Seiji Kajita and Tomoko Matsui, Spatial Statistics (2021), 43(3):100520; DOI: 10.1016/j.spasta.2021.100520
"Scalable Model Selection for Spatial Additive Mixed Modeling: Application to Crime Analysis", Daisuke Murakami, Mami Kajita, and Seiji Kajita, 2020, ISPRS Int. J. Geo-Inf. 2020, 9 (10), 577; DOI: 10.3390/ijgi9100577
"Crime prediction by data-driven Green’s function method", Mami Kajita and Seiji Kajita, 2019, International Journal of Forecasting Volume 36, Issue 2, April–June 2020, Pages 480-488, DOI:10.1016/j.ijforecast.2019.06.005
Patent
    Japanese Patent Application No. 2020-061937
    Japanese Patent Application No. 2019-518001, US, EU, CH, PCT
    Japanese Patent Application No. 2018-524169
Panels
(Materials available)
2019
2018
  • CEATEC Panel Discussion "Promotion of Open Data and Utilization of Open Data for the AI Era" 2018/10/19
  • Public-Private Roundtable "Crime Occurrence Information Related Data" 2018/03/29
  • Metropolitan Police Department Expert Study Group 2018/02/08
  • In addition, Tokyo Customs, Tokyo National Tax Bureau, Police Policy Society .... etc.