Lift Ratio
Confidence alone can be misleading when evaluating an association rule - it doesn't account for how popular the consequent item already is on its own. That's where the Lift Ratio comes in.
The Formula
Lift(A -> B) = Confidence(A -> B) / Support(B)
Lift compares the observed co-occurrence of A and B against what you'd expect if A and B were purchased completely independently of each other.
Interpreting Lift
| Lift Value | Meaning |
|---|---|
| Lift = 1 | A and B are independent - no real association |
| Lift > 1 | A and B occur together more often than expected - positive association |
| Lift < 1 | A and B occur together less often than expected - negative association |
Why Lift Matters
A rule can have high confidence simply because the consequent item is bought by almost everyone, regardless of the antecedent. Lift corrects for this by measuring the actual strength of the relationship rather than just how often the pattern shows up.
Implementing in Python
from mlxtend.frequent_patterns import apriori, association_rules
frequent_itemsets = apriori(df, min_support=0.2, use_colnames=True)
rules = association_rules(frequent_itemsets, metric="lift", min_threshold=1.0)
print(rules[["antecedents", "consequents", "confidence", "lift"]])
Coming Up Next
This wraps up Clustering, Dimensionality Reduction and Market Basket Analysis - rounding out the unsupervised learning portion of the course. Next, the course moves into Deep Learning, starting with the Perceptron, the simplest building block of a neural network.