Інформаційна технологія оптимізації параметрів ансамблю моделей штучного інтелекту для прогнозування наявності опадів за даними метеомоніторингу

Abstract

Data forecasting is a trivial task of systems analysis, there are different types of forecasts and predictions. One of them is a binary forecast that answers the question of whether an event will occur or not. One of the issues of meteorology is the issue of forecasting precipitation, as well as what accuracy will be in such a forecast. The paper considers the problem of forecasting the presence of precipitation according to meteorological monitoring and proposes information technology to optimize the parameters of the ensemble of such models of machine learning as models of gradient boosting and logistic regression, built on a set of informative features. The proposed information technology allows you to perform intelligence analysis of input data and determine the optimal set of informative features, and due to the algorithm, which at each step determines the optimal one, two, three,… -element sets of features that maximize fore-casting accuracy. Graphs of influence of signs on accuracy of the used models of machine learning are constructed. Each type of model has its own set of features. To provide information technology, the data provided by the Vinnytsia Center for Hydrometeorology were selected. These are the data of atmospheric monitoring of Vinnytsia for the last 10 years, which include: air temperature, humidity, dew point, cloudiness and wind speed. To increase the accuracy of forecasting, a mathematical model is proposed, which is based on the optimal determination of the weights of the ensemble of models of gradient boosting and logistic regression. Experiments were performed that showed a fairly accurate result. The accuracy of the proposed information technology showed 80%. This confirmed the reliability of the proposed technology.