Research on the application of time series in early risk detection systems for ІТ projects

Abstract

The article substantiates the relevance of implementing early risk detection systems in IT projects based on the analysis of time series formed from unstructured data of digital ecosystems and social media. The modern environment for implementing IT projects is characterised by high dynamism, non-linearity of information flows and a significant influence of subjective factors of stakeholders, which makes traditional monitoring methods insufficiently effective for identifying hidden threats. The authors define the role of time series as ‘leading indicators’ that allow the transformation of unstructured user responses (number of mentions, tone of messages, frequency of incidents) into formalised indicators. A detailed analysis of the mathematical apparatus for decomposing time series into trend, seasonal, cyclical, and random components is conducted. Particular attention is paid to the interpretation of the random component as a source of signals about the occurrence of significant management events and anomalies. The methodology for identifying models within the Box-Jenkins approach is considered. A description and visualisation of the autocorrelation (ACF) and partial autocorrelation (PACF) functions, which are key to determining the parameters of autoregression and moving average models, is provided. The use of ARIMA and SARIMA models for forming a baseline level of activity is justified. Deviations of actual data from this level are interpreted as early risk signals, which allows project management to move from reactive to proactive management.