UDC 614.849
This paper aims to improve the efficiency of fire departments' emergency response by optimizing the process of determining the required resources. The authors set the objective of formalizing this process as a regression problem in machine learning. Using facility characteristics (fire-tactical and incident-specific features), they must predict the quantitative parameters of forces and resources, including the number of firefighting and fire protection units, personnel, units, and fire trucks. For solvation of this problem the authors proposed a modified two-stage algorithm based on gradient boosting. In the first stage, an integral indicator introduced to reduce the dimensionality of the factor space and account for 12 heterogeneous fire-tactical characteristics of the facility. The weights for this indicator were calculated using the principal component analysis. In the second stage, a gradient boosting algorithm was implemented, using decision trees of a specified depth, ensuring that all features are taken into account, as weak predictors. The model was trained and tested on a sample of 1,507 scenarios. The forecast quality was assessed using the mean absolute error (MAE). The best results were achieved for predicting the number of fire trucks and GSP (errors of 3-14% of the average value); for other resources, the error was 20-30%. The developed approach enables automation and acceleration of decision-making during fire suppression, which will ultimately reduce time and material losses
likvidaciya pozhara, pozharno-takticheskie harakteristiki, metod glavnyh komponent, gradientnyy busting
1. Shemelova O.V. Metod "Dereva resheniy" pri reshenii zadach teorii veroyatnostey// Evraziyskoe Nauchnoe Ob'edinenie. 2021. No 12-1 (82). S. 43-45. EDN: https://elibrary.ru/INSOCX
2. Meftahetdinova D.R., Magerramov Z.T. Analiz dereva resheniy i logisticheskiy regressionnyy analiz kak analiticheskie metody sistemy podderzhki prinyatiya resheniy// Sciences of Europe. 2023. No 112 (112). S. 69-73. EDN: https://elibrary.ru/DITMUV
3. Anastasiadi D. E., Vereschagin A.A., Pyatakov M.A. Mashinnoe obuchenie i ego primenenie v razlichnyh oblastyah, vklyuchaya medicinu, finansy i proizvodstvo // Nauchnyy aspekt. 2024. T. 46. No4. S. 6166 -6174. EDN: https://elibrary.ru/EKASDD
4. Konovalov G.G. primenenie mashinnogo obucheniya dlya optimizacii zaprosov v sistemah upravleniya bazami dannyh // Mezhdunarodnyy zhurnal gumanitarnyh i estestvennyh nauk. 2023. No 10-2 (85). - S. 58-61. EDN: https://elibrary.ru/NHOWYU
5. Saltanaeva E.A., Shakirov A.A., Gimaeva A.R. Sravnenie tradicionnyh metodov mashinnogo obucheniya i glubokogo obucheniya// Nauchno-tehnicheskiy vestnik Povolzh'ya. 2023. No 12. S. 379-381. EDN: https://elibrary.ru/EQNUHN
6. Salimgareeva D.A. Metod glavnyh komponent// NovaInfo.Ru. 2017. T. 3. No 58. S. 5-9. EDN: https://elibrary.ru/XRETLZ
7. Fomina E.E. Primenenie kategorial'nogo metoda glavnyh komponent v prikladnyh sociologicheskih issledovaniyah// Russian Economic Bulletin. 2022. T. 5. No 1. S. 18-25. EDN: https://elibrary.ru/DCAJQS
8. Karter D. Iskusstvennyy intellekt. Mashinnoe obuchenie / D. Karter. - Moskva: "Avtor" 2024. 268 s.
9. Borisova L.R. Analiz prodolzhitel'nosti zhizni v regionah Rossiyskoy Federacii po metodam mashinnogo obucheniya i regressionnomu analizu// Sovremennaya matematika i koncepcii innovacionnogo matematicheskogo obrazovaniya. 2021. T. 8. No 1. S. 224-230. EDN: https://elibrary.ru/GKLHZL
