MLA-C01 Question 96
Select 4You are building a machine learning model to predict loan defaults. During data exploration, you notice that the dataset has significant class imbalance, with 'default' being much less represented than 'non-default'. Additionally, the dataset contains a mix of numerical, text, and image features. Which strategies can you use to address the class imbalance effectively during model training?
- A
Generate synthetic samples for the minority class using techniques like SMOTE for numerical features.
- B
Use data augmentation techniques such as flipping and cropping for the image data in the minority class.
- C
Downsample the majority class to match the size of the minority class to balance the dataset.
- D
Normalize the numerical features to ensure the features are on the same scale before training the model.
- E
Generate synthetic text samples using methods like text paraphrasing or language models for the minority class.
Show answer and explanation
Correct answers: A, B, C, E
Explanation
Class imbalance (CI) can lead to biased machine learning models. To address CI, specific strategies are required depending on the data type. For numerical data, synthetic data generation techniques like SMOTE can effectively increase the minority class. For image data, augmentation techniques such as flipping or cropping can improve class representation. For text data, synthetic generation methods like paraphrasing can be used. Downsampling the majority class is also a valid generic strategy. However, normalizing numerical features is unrelated to addressing CI.
- A. Correct.
This is correct. Synthetic Minority Oversampling Technique (SMOTE) is a widely used approach for generating synthetic samples for the minority class in numerical datasets.
- B. Correct.
This is correct. Data augmentation techniques such as flipping, cropping, or rotating images can effectively increase the representation of the minority class in image datasets.
- C. Correct.
This is correct. Downsampling the majority class is a valid strategy to balance the dataset, particularly when combined with other approaches like oversampling or synthetic data generation.
- D. Incorrect.
This is incorrect. Normalizing numerical features addresses scaling issues but does not directly address class imbalance in the dataset.
- E. Correct.
This is correct. Generating synthetic text samples using paraphrasing or language models can help address class imbalance in text datasets by increasing the representation of the minority class.