Russia Develops Program to Assess Individual Health Risks in the Workplace
The tool makes it possible to move from group-based assessments to personalized risk forecasting.
Russian researchers have developed a program that predicts occupational health risks for individual workers with an accuracy of 89%. The technology was created by scientists from Perm Polytechnic University together with specialists from Rospotrebnadzor and the Federal Risk Management Center.
According to a paper published in the scientific and practical journal Health Risk Analysis, existing risk assessment methods are usually based on group averages and fail to account for a number of important factors. Medical checkups, meanwhile, often identify health problems only after diseases have already developed.
At the core of the new system is an adaptive neuro-fuzzy network capable of working with incomplete and imprecise data. The program includes a mathematical model that is trained on diverse datasets and identifies complex patterns. A second component allows the system to interpret qualitative terms such as “high noise” or “moderate risk,” even when the boundaries of these concepts are not clearly defined.
How the System Was Trained
The system was trained using a database containing archival medical records of workers involved in underground copper-nickel mining. The dataset included 175,000 indicators. Of these, 80% were used for training and 20% for validation. During the initial phase, the neural network ran through 100 training cycles, searching for relationships between harmful workplace factors, a person’s health status, and the development of specific diseases. After each cycle, the program compared its predictions with real data, calculated the error rate, and automatically adjusted its internal parameters. Two graphs generated after each cycle helped monitor the training process.
Operating Principles and Testing
Academician of the Russian Academy of Sciences, Doctor of Medical Sciences, and Professor Nina Zaitseva provided further details on how the system works.
The model’s performance was tested on employees who had not been part of the training dataset, as well as on a fully independent sample consisting of data on drilling operators. In both cases, the program predicted occupational health risks with high accuracy.
