METHOD OF FORECASTING METROLOGICAL RISKS OF PRODUCT QUALITY

Abstract

The article considers the application of time series forecasting models for assessing the metrological risk of product quality at the stage of product manufacturing. Metrological risk, which is defined through such parameters as risk significance, probability of occurrence and probability of detection, acts as a time series that requires accurate forecasting for risk management in the production process. Six models are used in this paper to forecast metrological risk: Facebook Prophet, SARIMAX Statsmodels, Forecaster Recursive, Forecaster Direct, LGBMRegressor, and Linear Regression. Each model was tested on data covering the period from 1 January 2023 to 24 November 2024. The input data was divided into two sets: for training the neural network and for comparing the predicted values with the actual ones. The results of the study showed that the most accurate forecasts of metrological risk are provided by Facebook Prophet and SARIMAX Statsmodels, which demonstrated the best results in terms of the main accuracy metrics, including MAE, RMSE, MAPE, and R2. In addition, a comparison of the forecasting results visually demonstrated that these models are able to more accurately reflect real changes in risk over time. Logistic regression and other models, as shown by the results, have limited effectiveness in predicting metrological risk. Taking into account the results of the study, it can be argued that the use of time series models is an effective approach to predicting metrological risks in production processes. This allows not only to reduce the likelihood of unforeseen situations, but also to optimise product quality control. In the future, it is planned to improve the models to reduce the level of errors, increase their adaptability to changing production conditions and integrate forecasting models into decision-making systems.