Handling Business Specific Random and Cyclical Variation is key to a Robust Anomaly Detection Solution

For any business, monitoring key KPIs and metrics is important from strategic as well as tactical point of view. It gives you a complete overview of progress towards your goal and improve the decision making. Each business domain is different and hence the metrices for each of the domains behave differently. For example, in one of the deployments that we worked on for an e-commerce customer, we observed that every time there was a La Liga match, the traffic and usage on the app increased multi-fold in a region.

Business KPIs and metrics vary in terms of their behaviour, pattern and distribution and hence it is important to have compendium of models with each model capable of handling specific metrics based on nature of metrics and type of metrics (discrete, continuous, irregular, monotonic, multimodal, stationary, temporal etc.). The auto model selection feature within CrunchMetrics selects the most optimal model for each metrics that is subjected to anomaly detection and the auto hyperparameter tuning feature selects the most optimal parameter based on variation in data and selected model. We at CrunchMetrics have trained our proprietary anomaly detection models on cross industry datasets and bunch of empanelled datasets and hence the algorithms can handle any business specific random and cyclical variation effectively. The algorithms can detect structural change or trend breaks by fitting locally stationary autoregressive models and can effectively handle structural trend breaks by doing model selection using Minimum Description Length. Effects of day of the week, week of the month, day of the month, month of year, special event, holidays etc. can be determined and controlled. The algorithm is designed in a way to capture and handle temporal shift in the data. We tune our models’ parameter based on business specific random and cyclical variations. Our proprietary models self-evaluate and iterate over true targets and accuracy is evaluated using metrics like Scaled MAPE, MASE and quantile loss. Hence, the set of proprietary machine learning models within CrunchMetrics help accelerate business outcomes by handling business specific random and cyclical variations and thereby minimizing false positives and false negatives as we did in the case of the E-commerce deployment.

Summon the power of Augmented Analytics to help you identify risks and business incidents in real-time.

Sign up for a pilot run on your data

Leave a Reply

Your email address will not be published. Required fields are marked *