Anomaly Detection
- Anomaly Detection In order to catch the significant increase/decrease, even we think that the business doing ok/not well, the anomaly detection allow us the the alarming days or weeks or time periods. This process also allows us to see where the business did well and where can be improved.
-
How does Anomaly Detection work?
Anomaly Detection mainly concerns with the significant increase/decrease of a data point in the given metric. These metrics are Daily Funnel, Daily Cohort Anomaly, Daily Orders Anomaly, CLV Prediction of RFM Vs Current RFM Anomaly. Each metric of values of abnormal measurements is detected by AutoEncoder. AutoEncoder generates scores for each metric of the data point. Residuals (the difference between the actual value and predicted value) are calculated. The outliers of the residuals are the Abnormal values. -
Daily Funnel Anomaly
Daily Funnel is the actions of totals per day. With the combination of all actions (from session count to purchase count per day) it is possible to detect abnormal days by using ML techniques. -
Daily Cohort Anomaly & Daily Cohort Anomaly With Scores (Download to First Order)
Date columns represents downloaded day and each purchase count column represents the day that the customers had the first order after they have download. If there is a abnormal date which have significant low/high first purchase count related to downloaded date, this chart allow us the see the exact downloaded date as abnormal date. -
Daily Orders Anomaly
This chart allows us to see the % of increase/decrease compared with previous days of purchase counts. -
CLV RFM Vs Current RFM Anomaly
There are engaged customers whose purchases are predicted with CLV Prediction. We also know their Recency - Monetary - Frequency values that are calculated with their historic purchases. If we calculate their future RFM values via the CLV prediction and subtract them in order to detect a significant increase/decrease for each metric, we might clearly see how our customers of behaviors might change in the future.