Specialists from the Faculty of Computational Mathematics and Cybernetics of Moscow State University named after M. V. Lomonosov have created the Survivors data library, which helps to assess the risk of patient death and the probability of medical equipment failure.
The Survivors library is developed in Python and is designed for analyzing events over time. It takes into account complex dependencies in the data and works correctly even in the presence of missing values, which increases the accuracy of forecasts.
The versatility of the tool is noted. In medicine, it is used to predict risks and assess the life expectancy of patients based on rules understandable to experts. In CRM systems, Survivors calculates the probability of customer churn, and in industry it predicts equipment degradation and monitors the condition of systems.
Traditional survival analysis methods require strict assumptions about the distribution of events and complex data preparation. The new library uses machine learning and integrates into existing analytical systems, allowing you to avoid these limitations.
Survivors does not require data preprocessing and provides high forecast accuracy. The developers sought to create a tool capable of working with real and incomplete data, taking into account their features and uncertainty.
The library's algorithms are optimized for processing large datasets using parallel computing and efficient partitioning methods. Testing on nine open medical and industrial datasets showed stable and accurate forecasts, surpassing classical models with minimal settings.
Read materials on the topic:
Росатом запускает мощный центр обработки данных в Нижегородской области
Новую модель вычислительной инфраструктуры с ИИ создали в МГУ
Now on home
