Decision Tree Pruning Quiz

Overfitting occurs when the tree models the training data precisely, but the accuracy on test data may be worse. Pruning can help by improving accuracy initially, but too much pruning can worsen accuracy. Common pruning methods include setting max.tree depth and min.samples per node.

Association rule mining involves finding associations between items that occur at the same time or in the same context. It can be applied to shopping baskets to discover rules that can guide decisions, such as how to lay out supermarket shelves or make recommendations.

Pros of using K-nearest neighbor systems for classification and regression include good accuracy in many situations, explainable decisions, and the ability to display the basis of the decision. Cons include memory inefficiency, computational expense, and the need to compute distances to every data item to make each classification.

Market basket analysis is a technique used to identify relationships between products or items in a transactional dataset. A list of transactions is required for this analysis.

A frequent item-set is a group of items that occur together in a transaction dataset, and has a large count.

The Apriori algorithm is a heuristic algorithm that reduces the combinatorial size of the search space for frequent item-sets. It helps with association rule learning by identifying all the subsets that are frequent, based on the frequency of larger item-sets.

Association rule mining can be used to identify potential cross-selling opportunities, fraud detection, and insight into customer behavior, such as their preferences for product bundling or optional services.

Association rules can generate a large number of trivial and repetitive rules. Analysts can address this issue by setting minimum support and confidence thresholds, and making subjective decisions regarding the validity and importance of the rules.

Market basket analysis looks at item-sets in transactions, while association rule learning analyzes co-occurrences of items bought together or by the same users.

Heuristics are used to reduce the combinatorial size of the search space.

Lift determines whether LHS and RHS are dependent on each other, independent, or have a negative effect on each other.

The core concept of the Apriori algorithm is that if an item-set is frequent, then all of its subsets must also be frequent.

Applications of association mining include insights into customer behavior, cross-selling opportunities, and fraud detection.

A potential issue with association rules is that they can generate a huge number of trivial and repetitive rules.

Virtual items can expand the scope of association mining from products to any categorical variable of interest, such as store location or payment mode.

Market Basket Analysis requires a list of transactions, where items that occur together are called item-sets.

Association Rule Learning analyzes things that happen together, such as items bought together, viewed together, or bought by the same user on different days.

Algorithms like Apriori use heuristics to reduce the combinatorial size of the search space, such as specifying minimum rule support and rule confidence.

Association rules indicate co-occurrence, not causality or a sequence over time. Rule support is the proportion of transactions that contain the item set, while rule confidence is the proportion of rule firings that are correct predictions. The lift measures the dependency between the items in a rule, where a lift of 1 implies no association, a lift > 1 implies a positive dependency, and a lift < 1 implies a negative dependency.

The Apriori algorithm is a core concept in association mining, where if an item-set is frequent, then all of its subsets must also be frequent.

Association mining applications include insights into the next product a customer may buy, how to bundle optional services, possible cross-sells, and detecting fraud.

Issues with association rules include generating a large number of trivial and repetitive rules, requiring the analyst to make subjective decisions regarding validity and importance. Virtual items can expand the scope of association mining to any categorical variable of interest, such as store location or payment mode. Numerical concepts can be included in baskets by first binning them, such as age. Personalized recommendation approaches include collaborative filtering, which matches users to people with similar preferences, and content-based recommendation, which recommends based on what the user has bought or viewed in the past.