Customer

  • Major North American telecommunications company
  • Publicly traded company
  • Over $15 billion in revenue and 65,000 employees

Problem

  • Text messaging is a target for phishing, fraud and spamming techniques
  • Existing detection approach relied on legacy tactics and static rules
  • Inability to adapt to new spamming techniques
  • Large resource costs from manual review of flagged messages

Solution

  • Created an Machine Learning model using the Splunk MLTK
  • Used a classification approach for predicting SMS spam
  • Observations/events were categorized into discrete groups

Result

  • The ML model found to be 93% accurate, significantly reducing false positives
  • Greatly reduced cost of internal resources required for manual checks
  • Model is continually learning and improving from data in an unsupervised manor
  • Removed all need for static rules and thresholds