Con dence and lift

M A R K E T B A S K E T A N A LY S I S I N P Y T H O N

Isaiah Hull
Economist
When support is misleading
TID Transaction  

1 Coffee, Milk 1. Milk and bread frequently purchased

together.
2 Bread, Milk, Orange Support: {Milk} → {Bread}
3 Bread, Milk 2. Rule is not informative for marketing.
Milk and bread are both popular items.
4 Bread, Milk, Sugar

5 Bread, Jam, Milk

... ...

MARKET BASKET ANALYSIS IN PYTHON

The con dence metric
1. Can improve over support with additional metrics.

2. Adding con dence provides a more complete picture.

Support(X&Y )
Support(X)

MARKET BASKET ANALYSIS IN PYTHON

Interpreting the con dence metric

Support(Milk&Coffee) = 0.20 Support(Milk) = 1.00

MARKET BASKET ANALYSIS IN PYTHON

Interpreting the con dence metric
 
Support(Milk&Coffee) 0.20
= = 0.20
Support(Milk) 1.00

 
Support(Milk&Coffee) 0.20
= = 0.25
Support(Milk) 0.80
 
Support(Milk&Coffee) 0.20
= = 1.00
Support(Milk) 0.20

MARKET BASKET ANALYSIS IN PYTHON

The lift metric
 

Lift provides another metric for evaluating the relationship between items.
Numerator: Proportion of transactions that contain X and Y.

Denominator: Proportion if X and Y assigned randomly and independently.

Support(X&Y )
Support(X)Support(Y )

MARKET BASKET ANALYSIS IN PYTHON

Preparing the data
from mlxtend.preprocessing import TransactionEncoder
import pandas as pd

# Split library strings into lists

libraries = data['Library'].apply(lambda t: t.split(','))

# Convert to list of lists

libraries = list(libraries)

# One-hot encode books

books = TransactionEncoder().fit(libraries).transform(libraries)

# Convert one-hot encoded data to DataFrame

books = pd.DataFrame(books, columns = encoder.columns_)

MARKET BASKET ANALYSIS IN PYTHON

Computing con dence and lift
# Print first five items
print(books.head())

Hunger Gatsby
0 False True
1 False True
2 False False
3 False True
4 False True

Dataset: GoodBooks-10K.

MARKET BASKET ANALYSIS IN PYTHON

Computing con dence and lift
# Computing support.
supportHG = np.logical_and(books['Hunger'],books['Gatsby']).mean()
supportH = books['Hunger'].mean()
supportG = books['Gatsby'].mean()

# Compute and print confidence and lift.

confidence = supportHG / supportH
lift = supportHG / (supportH * supportG)

# Print results.
print(supportG, confidence, lift)

(0.30, 0.16, 0.53)

MARKET BASKET ANALYSIS IN PYTHON

 Let's practice!
M A R K E T B A S K E T A N A LY S I S I N P Y T H O N
 Leverage and
conviction
M A R K E T B A S K E T A N A LY S I S I N P Y T H O N

Isaiah Hull
Economist
Building on simpler metrics
F requency(X)
Support(X) =
N
F requency(X&Y )
Support(X → Y ) =
N
Support(X → Y )
Confidence(X → Y ) =
Support(X)
Support(X → Y )
Lift(X → Y ) =
Support(X)Support(Y )

MARKET BASKET ANALYSIS IN PYTHON

The leverage metric
Leverage also builds on support.

Leverage(X → Y ) =

Support(X&Y ) − Support(X)Support(Y )

Leverage is similar to lift, but easier to interpret.

Leverage lies in -1 and +1 range.

Lift ranges from 0 to in nity.

MARKET BASKET ANALYSIS IN PYTHON

Computing leverage
# Compute support for Twilight and Harry Potter
supportTP = np.logical_and(books['Twilight'], books['Potter']).mean()

# Compute support for Twilight

supportT = books['Twilight'].mean()

# Compute support for Harry Potter

supportP = books['Potter'].mean()

# Compute and print leverage

leverage = supportTP - supportP * supportT
print(leverage)

0.018

MARKET BASKET ANALYSIS IN PYTHON

The conviction metric
1. Conviction is also built using support.

2. More complicated and less intuitive than leverage.

Conviction(X → Y ) =

Support(X)Support(Ȳ )
Support(X&Ȳ )

MARKET BASKET ANALYSIS IN PYTHON

 Interpreting conviction

1 Images taken from goodreads.com.

MARKET BASKET ANALYSIS IN PYTHON

Computing conviction
# Compute support for Twilight and Harry Potter and Twilight
supportTP = np.logical_and(books['Twilight'], books['Potter']).mean()
supportT = books['Twilight'].mean()

# Compute support for NOT Harry Potter

supportnP = 1.0 - books['Potter'].mean()

# Compute support for Twilight and NOT Harry Potter

supportTnP = supportT - supportPT

# Compute conviction
conviction = supportT*supportnP / supportTnP
print(conviction)

1.16

MARKET BASKET ANALYSIS IN PYTHON

 Let's practice!
M A R K E T B A S K E T A N A LY S I S I N P Y T H O N
 Association and
dissociation
M A R K E T B A S K E T A N A LY S I S I N P Y T H O N

Isaiah Hull
Economist
 Using dissociation to pair ebooks

1 Images taken from goodreads.com.

MARKET BASKET ANALYSIS IN PYTHON

 Introduction to Zhang's metric
1. Introduced by Zhang (2000)
Takes values between -1 and +1

Value of +1 indicates perfect association

Value of -1 indicates perfect dissociation

2. Comprehensive and interpretable

3. Constructed using support

1 Zhang, T. (2000). Association Rules. Proceedings of the 4th Paci c-Asia conference, PADKK, pp.245-256. Kyoto,
Japan.

MARKET BASKET ANALYSIS IN PYTHON

De ning Zhang's metric
 
Zhang(A → B) =

Confidence(A → B) − Confidence(Ā → B)
Max(Confidence(A → B), Confidence(Ā → B))

 
Support(A&B)
Confidence =
Support(A)

MARKET BASKET ANALYSIS IN PYTHON

Constructing Zhang's metric using support
 
Zhang(A → B) =
Support(A&B) − Support(A)Support(B)
Max[Support(AB)(1 − Support(A)), Support(A)(Support(B) − Support(AB)]

MARKET BASKET ANALYSIS IN PYTHON

Computing Zhang's metric
# Compute the support of each book
supportH = hobbit.mean()
supportP = pride.mean()

# Compute the support of both books

supportHP = np.logical_and(hobbit, pride).mean()

MARKET BASKET ANALYSIS IN PYTHON

Computing Zhang's metric
# Compute the numerator
num = supportHP - supportH*supportP

# Compute the denominator

denom = max(supportHP*(1-supportH), supportH*(supportP-supportHP))

# Compute Zhang's metric

zhang = num / denom
print(zhang)

0.08903

MARKET BASKET ANALYSIS IN PYTHON

 Let's practice!
M A R K E T B A S K E T A N A LY S I S I N P Y T H O N
 Advanced rules
M A R K E T B A S K E T A N A LY S I S I N P Y T H O N

Isaiah Hull
Economist
Overview of market basket analysis
Standard procedure for market basket analysis.
1. Generate large set of rules.

2. Filter rules using metrics.

3. Apply intuition and common sense.

MARKET BASKET ANALYSIS IN PYTHON

Generating rules
Number of rules grows exponentially in number of items.
Most rules are not useful.

Must apply initial round of ltering.

Covered in chapter 3 using Apriori algorithm

MARKET BASKET ANALYSIS IN PYTHON

How does ltering work?
ID antecedents consequents support Lift

1 Harry Potter The Hunger Games 0.001 1.01

2 Hunger Games Twilight 0.020 1.23

3 Pride and Prejudice The Hobbit 0.030 1.05

4 The Hobbit Twilight 0.015 0.85

5 Harry Potter The Hobbit 0.010 1.07

MARKET BASKET ANALYSIS IN PYTHON

Multi-metric ltering
ID antecedents consequents support zhang

1 Harry Potter The Hunger Games 0.001 -0.05

2 Hunger Games Twilight 0.020 0.17

3 Pride and Prejudice The Hobbit 0.030 0.06

4 The Hobbit Twilight 0.015 -0.04

5 Harry Potter The Hobbit 0.010 0.34

MARKET BASKET ANALYSIS IN PYTHON

Performing multi-metric ltering
print(rules.head())

antecedents consequents antecedent support ... support confidence ... conviction

0 Potter Hunger 0.48 ... 0.12 ... 0.26 ... 0.92
1 Potter Hunger 0.32 ... 0.12 ... 0.26 ... 0.92
2 Twilight Hunger 0.26 ... 0.09 ... 0.35 ... 1.04
3 Hunger Twilight 0.32 ... 0.09 ... 0.28 ... 1.03
4 Mockingbird Hunger 0.48 ... 0.10 ... 0.20 ... 0.85

MARKET BASKET ANALYSIS IN PYTHON

Performing multi-metric ltering
# Select subset of rules with low consequent support.
rules = rules[rules['consequent support'] < 0.05]
print(len(rules))

12

# Select subset of rules with lift > 1.5.

rules = rules[rules['lift'] > 1.5]
print(len(rules))

MARKET BASKET ANALYSIS IN PYTHON

 Let's practice!
M A R K E T B A S K E T A N A LY S I S I N P Y T H O N