Preparing for Your Professional Machine Learning Engineer Journey

Google Cloud PMLE（ Professional Machine Learning Engineer ）專業機器學習工程師認證考試

Sample Question 130 / Question ID: 130

Question / 題目

You are developing a model to detect fraudulent credit card transactions. You need to prioritize detection, because missing even one fraudulent transaction could severely impact the credit card holder. You used AutoML to train a model on users' profile information and credit card transaction data. After training the initial model, you notice that the model is failing to detect many fraudulent transactions. How should you adjust the training parameters in AutoML to improve model performance? (Choose two.)

您正在開發一個用於檢測信用卡欺詐交易的模型。您需要優先考慮檢測率，因為漏掉任何一筆欺詐交易都可能對持卡人造成嚴重影響。您使用 AutoML 基於用戶的個人資料信息和信用卡交易數據訓練了一個模型。在訓練初始模型後，您發現該模型未能檢測出許多欺詐交易。您應該如何調整 AutoML 中的訓練參數以提高模型性能？（選擇兩項。）

A: Increase the score threshold

A: 提高分數閾值（門檻）

B: Decrease the score threshold

B: 降低分數閾值（門檻）

C: Add more positive examples to the training set

C: 在訓練集中增加更多正例（欺詐樣本）

D: Add more negative examples to the training set

D: 在訓練集中增加更多負例（正常樣本）

Analysis / 解析

A: Wrong. Increasing the score threshold (the confidence level required to classify a transaction as fraudulent) means the model will become more conservative and only flag transactions it is extremely certain about. This would decrease the number of detected frauds, thereby increasing false negatives (missing more fraudulent transactions), which directly contradicts the requirement.

A: 錯誤。 提高分數閾值（將交易分類為欺詐所需的置信度水平）意味著模型會變得更加保守，僅標記其極其確定的交易。這將減少檢測到的欺詐數量，從而增加偽陰性/漏報（錯失更多欺詐交易），這與題目需求直接矛盾。

B: Correct. Decreasing the score threshold makes the model more sensitive. It lowers the bar for a transaction to be classified as fraudulent, ensuring that more potential frauds are flagged. While this may increase false positives (flagging legitimate transactions as fraud), it significantly reduces false negatives (missing frauds), which aligns with the priority of protecting the credit card holder at all costs.

B: 正確。 降低分數閾值會使模型更加敏感。它降低了將交易歸類為欺詐的門檻，從而確保標記出更多潛在的欺詐行為。雖然這可能會增加偽陽性/誤報（將合法交易標記為欺詐），但它顯著減少了偽陰性/漏報（漏掉欺詐），這符合不惜一切代價保護信用卡持卡人的優先原則。

C: Correct. In credit card fraud detection, the dataset is typically highly imbalanced, meaning there are very few fraudulent transactions (positive examples) compared to legitimate ones (negative examples). Because the model is currently failing to detect many fraudulent transactions, adding more positive examples will help the model better learn the patterns and features associated with fraud, thereby improving its detection rate (Recall).

C: 正確。 在信用卡欺詐檢測中，數據集通常高度不平衡，這意味著與合法交易（負例）相比，欺詐交易（正例）非常少。由於該模型目前未能檢測出許多欺詐交易，因此增加更多正例將有助於模型更好地學習與欺詐相關的模式和特徵，從而提高其檢測率（召回率）。

D: Wrong. Adding more negative examples (legitimate transactions) to an already imbalanced dataset would exacerbate the class imbalance problem. The model would become even more biased toward predicting transactions as legitimate, leading to an even higher rate of missed fraudulent transactions.

D: 錯誤。 在本已不平衡的數據集中增加更多負例（合法交易）會加劇類別不平衡問題。模型將變得更加偏向於將交易預測為合法，從而導致錯失欺詐交易的機率更高。