Updated notebook

Author: paul-dave80

jupyter/anomaly_detection_creditcard.ipynb

@@ -74,6 +74,52 @@
     "Each step improves accuracy and reduces false alerts.\n",
     "\n",
     "\n",
+    "## Feature Processing\n",
+    "\n",
+    "The dataset includes processed features (V1–V28) created using statistical methods.\n",
+    "\n",
+    "These features help detect patterns but do not directly represent real-world transaction details. Because of this, extreme values can strongly influence model decisions.\n",
+    "\n",
+    "Careful handling of these values is important to reduce false positives.\n",
+    "\n",
+    "\n",
+    "## Business Impact\n",
+    "\n",
+    "### False Positives vs Missed Fraud\n",
+    "\n",
+    "- False positives → customer frustration and lost transactions  \n",
+    "- Missed fraud → financial loss and security risk  \n",
+    "\n",
+    "The goal is to balance both.\n",
+    "\n",
+    "\n",
+    "### Risks\n",
+    "\n",
+    "- Model may flag unusual but valid transactions  \n",
+    "- Data changes over time may reduce accuracy  \n",
+    "- Data adjustments may introduce bias if not reviewed  \n",
+    "\n",
+    "\n",
+    "### Recommendations\n",
+    "\n",
+    "- Improve data quality by handling extreme values  \n",
+    "- Monitor model performance continuously  \n",
+    "- Deploy changes gradually  \n",
+    "\n",
+    "\n",
+    "### Stakeholder Communication\n",
+    "\n",
+    "- Share regular updates  \n",
+    "- Clearly explain limitations  \n",
+    "- Set realistic expectations  \n",
+    "\n",
+    "## Conclusion\n",
+    "\n",
+    "This system provides a strong starting point for fraud detection using machine learning.\n",
+    "\n",
+    "While the current model has limitations, especially with false positives, it demonstrates how data-driven approaches can improve fraud detection.\n",
+    "\n",
+    "Ongoing improvements in data quality, model tuning, and monitoring will be key to long-term success.\n",
     "\n",
     "\n",
     "### About Jupyter Notebooks\n",