Using machine learning methods to predict the detection of crimes based on primary accounting documents

The result of the solving of crimes is one of the important indicators of the activities of law enforcement agencies. Despite the improvement of crime investigation methods, the success rate of crime detection in the Russian Federation remains at the level of 51%–56%. The article describes a method for constructing a mathematical model – a digital double of a registered crime. As the initial data for constructing the model, an array of information – primary accounting documents, about 341 thousand crimes committed on the territory of the Primorsky Krai over 11 years-from 2010 to 2020. The model allows you: with 88% confidence, based on the formalized primary information contained in the primary accounting documents – statistical cards Form No. 1 “On the detected crime”, to make a forecast about whether the crime will be solved or not; to audit unsolved crimes of previous years in order to determine the crimes that have a high probability of detection; to identify the features in the statistical cards that most affect the forecast of the detection of crimes. The model is based on the use of machine learning algorithms “gradient boosting over decision trees”, implemented in the open library of artificial intelligence CatBoost from Yandex. The accuracy of the model is confirmed by the preparation and verification of the forecast of the result of the investigation of crimes in January–June 2021 for 16408 crimes committed on the territory of the Primorsky Krai.

1. The form of the federal statistical monitoring No. 4-EGS “On the state of crime and the results of the investigation of crimes”. Available at: http://crimestat.ru/analytics (accessed 20.05.2021). (In Russ)

2. Nizamov V.Y. On the question of the concept of “crime disclosure” in criminalistics and criminal procedure. Leningrad legal journal. 2016;1(43):170-179. (In Russ)

3. Order of the General Prosecutor's Office of the Russian Federation, the Ministry of Internal Affairs of Russia, the Ministry of Emergency Situations of Russia, the Ministry of Justice of Russia, the Federal Security Service of Russia, the Ministry of Economic Development of Russia, the Federal Drug Control Service of Russia dated December 29, 2005 No. 39/1070/1021/253/780/353/399 “About the unified accounting of crimes”. Available at: https://rg.ru/2006/01/25/uchet-prestupleniy-dok.html (accessed: 20.05.2021). (In Russ)

4. Ovchinsky V.S., Larina E.S. Artificial intelligence: Big Data. Crime. – Moscow: Book World, 2018:1-416. (In Russ)

5. Yasnitsky L.N., Vauleva S.V., Safonova D.N., Cherepanov F.M. The use of artificial intelligence methods in the study of the personality of serial killers. Criminological journal of the Baikal State University of Economics and Law. 2015;9(3):423-430. (In Russ)

6. Pyankov D.D., Malyugin M.I., Yasnitsky L.N. The use of neural network technologies in the study of factors affecting crime in Russian cities. Artificial intelligence in solving urgent social and economic problems of the XXI century, May 21-23, 2019. Perm: Perm State National Research University. 2019:126-132. (In Russ)

7. Poponina A.O. Forecasting the crime rate in the regions of Russia. Artificial intelligence in solving urgent social and economic problems of the XXI century, May 21-23, 2019. Perm: Perm State National Research University, 2019:119-125. (In Russ)

8. Bulgakov D.Y. Modern Approaches to Testing Biometric Identification Systems Based on Facial Images. Artificial Intelligence (Big Data) in The Service of The Police. Moscow: Management Academy of the Ministry of the Interior of Russia, 2020:45-51. (In Russ)

9. Gordeev A.Y. Prospects for the development and use of artificial intelligence and neural networks to counter crime in Russia (based on foreign experience). Scientific portal of the Russian Ministry of internal Affairs. 2021;1(53):123-135. (In Russ)

10. Prokhorenkova L., Gusev G., Vorobev A., Dorogush A. V., Gulin A. Catboost unbiased boosting with categorical features. 2017. arXiv: 1706.09516.

11. Melnikov A.V., Narushev I.R., Kubasov I.A. Method for Evaluating Inhomogeneous Alternatives with the Hierarchical Structure of Unrelated Criteria Based on Medium-Consistent Matrix of Pair Comparisons. Journal of Computational and Engineering Mathematics. 2019;6(2):32-41. DOI: 10.14529/jcem190203.

12. Salakhutdinova K.I., Lebedev I.S., Krivtsova I.E. Algorithm of gradient boosting of decision trees in the problem of software identification. Scientific and technical bulletin of information technologies, mechanics and optics. 2018:1016-1022. DOI: 10.17586/2226-1494-2018-18-6-1016-1022. (In Russ)

13. Sokolov E. Seminars on the choice of models. 2015:1-9. Available at: http://www.machinelearning.ru/wiki/images/1/1c/Sem06_metrics.pdf (accessed: 20.05.2021). (In Russ)