Algoritmi riskiperusteisen valvonnan tukena: Pelastustoimen valvontakohteiden luokittelu ja valvontavälien muodostaminen

Abstract

The report presents a risk-based supervision interval model developed to support supervision planning and the targeting of fire inspections within the rescue services. The model was built during the development project Classification of Supervised Targets and Supervision Intervals between 1 May 2025 and 31 May 2026. The objective of the project was to establish a nationally more uniform and transparent method for determining indicative supervision intervals for supervised targets, allowing supervision to be targeted more consistently than before based on the risk level of the targets and the available supervision capacity. In the model, the risk of supervised premises is assessed by combining ignition probability, the consequence value of a fire, and an expert risk score derived from an expert survey. Ignition probability is assessed using data on building fires and building fire hazards, building stock data, and, where applicable, the floor-area dependency of ignition frequency. The consequence assessment takes into account property damage and personal injuries, expressed in commensurable monetary values. The expert risk score was derived from ranked responses in the expert survey using a pairwise comparison model, so that experience-based knowledge relevant to supervision can be distinguished from the statistical risk component. The risk components are combined into an overall risk score, which is converted into an indicative inspection interval using a capacity-constrained algorithm. The algorithm does not determine inspection intervals solely on the basis of risk ranking, but relates inspection frequencies to the total available supervisory capacity. Based on sensitivity analysis, the optimized contrast parameter of the algorithm produced an inspection interval distribution in which differences in risk are reflected in the inspection intervals, while avoiding excessive clustering of premises at the extreme ends of the inspection interval scale. The report also describes the refinements made on the basis of consultation feedback, the use of the model in the supervision planning of rescue departments, the positioning of the algorithm within rescue services’ knowledge-based management and the OEJ system, and the ethical boundary conditions of algorithmic analytics. The model is not intended to function as an automatic decision rule from which deviations are not permitted, but as an indicative starting point for supervision planning. In its final application, account must be taken of site-specific information, supervision history, local risk information, the effectiveness of supervision, and the documented discretion of the competent authority. The key conclusion of the project is that the allocation of rescue authorities’ supervisory resources can be supported by a transparent, reproducible risk model that takes capacity constraints into account. The strength of the model lies in the separation of the components of risk, the traceability of the calculation, and the fact that it makes the grounds for determining inspection intervals more visible than before. The limitations of the model relate in particular to classification correspondences between datasets, the quality of the data used, the statistical uncertainty associated with rare personal injuries, the dataset-dependence of expert judgment, and the fact that the model does not yet directly measure the effectiveness of supervision. In further development, the model can be supplemented with supervision history, a site-specific risk score, information system integration and, at a later stage, predictive analytics, when the quality of the datasets and the ethical governance mechanisms make this possible.