Credit Card Fraud Detection

Train the model to detect the credit card if fraud or not with supervised model and unsupervised model

Model

Supervised

• Encoder-Decoder + Classifier: the main idea here is try to transform the origin dataset to a different dimension space let the dataset becomes "linear separable"

  

Unsupervised

• Encoder-Decoder

Evaluation

Supervised

• Confusion matrix

  

• Score

  • F1 Score : 0.95
  • Fraud Recall : 0.915

Unsupervised

* 評估方法, 步驟如下 : 

1. 將測試資料進行模型預測
2. 計算原始資料與預測後的資料計算均方誤差(MSE)
3. 計算所有資料的MSE的平均值與標準差
4. 將落於均值正負1倍標準差外的資料視為異常值
5. 與真實標籤做評估

• 一倍標準差外的資料圖

  

• Confusion matrix

  

• Score

  • F1 Score : 0.86
  • Fraud Recall : 0.64

API