Imbalanced Data in Classification

The video tutorial discusses the challenges of imbalanced data in classification problems and explores various techniques to address this issue. It covers under sampling and over sampling methods, including advanced techniques like SMOTE and its variants. The tutorial also introduces adaptive synthetic sampling, which focuses on generating synthetic data for difficult-to-classify observations. These methods aim to improve the representation of minority classes in datasets, enhancing the performance of machine learning models.

What is the main issue with imbalanced data sets in classification tasks?

• They make the model too complex.

• They cause the model to favor the majority class.

• They require more computational power.

• They lead to overfitting.

What is a potential drawback of under-sampling?

• It requires more data storage.

• It can lead to overfitting.

• It may result in loss of important data.

• It increases the complexity of the model.

How does over-sampling address the imbalance in data?

• By duplicating existing minority class samples.

• By adding noise to the data.

• By changing the class labels.

• By removing majority class samples.

What is a drawback of simple over-sampling?

• It introduces new data points.

• It requires complex algorithms.

• It can lead to overfitting by repeating the same data.

• It reduces the size of the dataset.

What is the purpose of Tomek links in under-sampling?

• To add more data points.

• To remove noise from the data.

• To identify and remove overlapping data points.

• To increase the number of majority class samples.

What does SMOTE stand for?

• Synthetic Minority Over-sampling Technique

• Simple Minority Over-sampling Technique

• Statistical Minority Over-sampling Technique

• Standard Minority Over-sampling Technique

How does SMOTE generate new samples?

• By creating synthetic samples using linear interpolation.

• By duplicating existing samples.

• By removing majority class samples.

• By adding random noise to the data.

What is a limitation of SMOTE?

• It requires a lot of computational power.

• It can create linear patterns in the data.

• It reduces the dataset size.

• It only works with numerical data.

What is the main advantage of adaptive synthetic sampling over SMOTE?

• It is easier to implement.

• It requires less data.

• It gives higher weight to difficult-to-classify samples.

• It is faster to compute.

What is a common feature of SMOTE variants?

• They all require labeled data.

• They all reduce the dataset size.

• They all use nearest neighbors for generating samples.

• They all use Tomek links.