You want to train a model for detecting stones that contain gold. A mining company has a dataset of 5 million stone images, of which only 500 contain gold. What is the best measure to evaluate your model if you designate stone images with gold as the positive class?

You have a dataset of 5,000 audio files, and you want to train a model to detect music files within the dataset. How would you validate the performance of your classifier?

You are given a dataset of cars and asked to train a decision tree to classify the cars into two classes: Luxury and Sports. If a leaf node contains 100 cars, which case would provide better performance? (L represents the Luxury class, and S represents the Sports class.)

What is an underlying assumption of Naive Bayes Classification?

You want to train a model for detecting bald eagles. National Geography provides you a dataset of 1 million bird images, of which only 200 are bald eagles. What is the best measure to evaluate your model if you designate bald eagles as the positive class?

Compute the cosine similarity and correlation coefficient between feature vectors X and Y: X = (4, 2, 3, 3) Y = (1, 3, 5, 7)

You are given a dataset of agricultural products and asked to train a decision tree to classify the products into two classes: Fruits and Vegetables. If a leaf node contains 100 products, which case would provide better performance? (F represents the Fruits class, and V represents the Vegetables class.)

Your company recently introduced a new service, and you decide to conduct a survey among its 30 million users to get their feedback on the new service. Based on data mining techniques, which users will you survey?

Figure 1: Training Dataset for Car Ownership Instance Employed Insured Marital Status Car Ownership 1 Yes No Single Yes 2 Yes Yes Single No 3 No No Married Yes 4 No Yes Single Yes 5 Yes Yes Married No 6 No No Single No   Review the table labeled Figure 1: Training Dataset for Car Ownership. Assume we want to use a decision tree to predict if a person owns a car or not. Using entropy as the measure of node impurity, what is the information gain if the split is done on the attribute of being insured?

You have a dataset that contains information about different countries in the world. For each country, more than 20 attributes, such as population, area, weather, gross income, and so on, are available. What would NOT be a possible issue for learning models on this dataset?