Development of Anomaly Detection System, with the aim of issuing alerts in real time to detect suspicious transactions.
A single Neural Networks model is trained with Tensorflow, capable of detecting the patterns of multiple time series and making predictions.
From the past prediction errors, the confidence intervals are calculated for the different prediction horizons, which allows us to build the thresholds from which alerts will be issued.
Different thresholds are calculated, which will allow the alerts to be issued depending on the degree of deviation thereof.
All integrations are made with different messaging services to issue the corresponding alerts.