Abstract
The paper describes the joint application of text comparison model, time series anomaly detection and failure modes and effects analysis (FMEA) to calculate 3 software failure risk scores. The aim of the research is to develop and verify the software failure risk assessment model. The objectives are model development and experimentation. The study allows examining commonly used models from the perspective of their joint application in risk assessments.The study is based on a data set that contains: a list of 24-hour application execution log records with timestamps, severity levels and messages obtained from the Loki log database, a time series for the same period extracted from the Prometheus time series database. To create an initial dataset the test application has been developed which simulates RAM spikes with an error level entry added to the log. The experiment was performed using Kotlin and R languages. The study used Apache Commons Text package to perform text comparison. Time series is passed to the R runtime using the forecast library, a model is formed to detect anomalies and predict "normal" values at outlier points. Thus, a list of points on the time series with anomalous values is obtained and they are associated with log records. As a result of the experiment, a time series graph was obtained with the identified anomalous values and with "normal" values at these points.The final result presents 2 errors: experimental and non-experimental. The experimental error was rated occurrence of 1, detection of 1, severity of 3 and a final rating of 3. The non-experimental error was rated occurrence of 1, detection of 1, severity of 0 and a final rating of 0. It follows that the experimental error was rated non-zero and the non-experimental error was rated as "noise".