Association rule learning is a machine learning technique used to identify patterns, associations, and relationships between variables in large datasets. It is often applied in market basket analysis, where the goal is to discover which products are frequently bought together by customers. In market basket analysis, confidence, support, and lift are used to measure the strength of association between items purchased by customers.
-
Confidence: Confidence in market basket analysis measures the conditional probability of a product Y being purchased given that product X has already been purchased. A high confidence value indicates that customers who bought item X are likely to also buy item Y.
-
Support: Support measures the frequency of occurrence of an item or a set of items in all transactions. A high support value indicates that the item or set of items is frequently bought together by customers.
-
Lift: Lift measures the strength of association between two items by comparing the observed frequency of co-occurrence of both items with the expected frequency of co-occurrence under the assumption that the items are independent. A lift value greater than 1 indicates that the items are positively correlated, a lift value of 1 indicates no correlation, and a lift value less than 1 indicates that they are negatively correlated. A high lift value indicates that the two items are likely to be bought together, and the lift value can be used to identify strong association rules between items.
Two approaches of Asscoaite rule learning, known as appriori and FP growth, have been used to do a market basket analysis. My code is very simple to understand and can be extended by more analysis.
## data of this code can be downloaded from the following link:
#https://www.kaggle.com/datasets/heeraldedhia/groceries-dataset/code?datasetId=877335&sortBy=voteCount
import pandas as pd
import matplotlib.pyplot as plt
import numpy as np
from mlxtend.frequent_patterns import apriori, association_rules
from mlxtend.frequent_patterns import fpgrowth#reading data
dataset = pd.read_csv("Groceries_dataset.csv")
#Member_number is unique for each customer. Date is date of the transaction,
#itemDescription is the product bought for this date.
print(dataset.head())
print(dataset.shape)
#################################
#checking missing values
nan_values = dataset.isna().sum()
print(nan_values) Member_number Date itemDescription
0 1808 21-07-2015 tropical fruit
1 2552 05-01-2015 whole milk
2 2300 19-09-2015 pip fruit
3 1187 12-12-2015 other vegetables
4 3037 01-02-2015 whole milk
(38765, 3)
Member_number 0
Date 0
itemDescription 0
dtype: int64
##Basket analysis: we can see the bought products of clients for everyday
client_basket=dataset.groupby(['Member_number','Date'])['itemDescription'].apply(sum)
print(client_basket)
## we can see the bought products of clients for every day withot client number
clinet_basket2 = [a[1]['itemDescription'].tolist() for a in list(dataset.groupby(['Member_number','Date']))]
print(clinet_basket2[0:10])Member_number Date
1000 15-03-2015 sausagewhole milksemi-finished breadyogurt
24-06-2014 whole milkpastrysalty snack
24-07-2015 canned beermisc. beverages
25-11-2015 sausagehygiene articles
27-05-2015 sodapickled vegetables
...
4999 24-01-2015 tropical fruitberriesother vegetablesyogurtkit...
26-12-2015 bottled waterherbs
5000 09-03-2014 fruit/vegetable juiceonions
10-02-2015 sodaroot vegetablessemi-finished bread
16-11-2014 bottled beerother vegetables
Name: itemDescription, Length: 14963, dtype: object
[['sausage', 'whole milk', 'semi-finished bread', 'yogurt'], ['whole milk', 'pastry', 'salty snack'], ['canned beer', 'misc. beverages'], ['sausage', 'hygiene articles'], ['soda', 'pickled vegetables'], ['frankfurter', 'curd'], ['sausage', 'whole milk', 'rolls/buns'], ['whole milk', 'soda'], ['beef', 'white bread'], ['frankfurter', 'soda', 'whipped/sour cream']]
#Converting Date into DateTime type
Date=dataset.set_index(['Date'])
Date.index=pd.to_datetime(Date.index, infer_datetime_format= True)## Data analysis
#The items sold per month, week, and day have been plotted to see the variations and the average level of sale.
fig1 = plt.figure("Figure 1")
ax = plt.axes()
ax.set_facecolor('silver')
Date.resample("D")['itemDescription'].count().plot(figsize=(12,5), grid=True,
color='black').set(xlabel="Date", ylabel="Total Number of Items Sold")
Date.resample("W")['itemDescription'].count().plot(figsize=(12,5), grid=True,
color='blue').set(xlabel="Date", ylabel="Total Number of Items Sold")
Date.resample("M")['itemDescription'].count().plot(figsize=(12,5), grid=True,
color='red',title="Items Sold per month in red, per week in blue and per day in black").set(xlabel="Date", ylabel="Total Number of Items Sold")[Text(0.5, 0, 'Date'), Text(0, 0.5, 'Total Number of Items Sold')]
#The number of customers per month, week, and day has been plotted to see the variations and the average number of customers.
fig2 = plt.figure("Figure 2")
ax = plt.axes()
ax.set_facecolor('silver')
Date.resample('D')['Member_number'].nunique().plot(figsize=(12,5), grid=True,
color='black').set(xlabel="Date", ylabel="Number of customers")
Date.resample('W')['Member_number'].nunique().plot(figsize=(12,5), grid=True,
color='blue').set(xlabel="Date", ylabel="Number of customers")
Date.resample('M')['Member_number'].nunique().plot(figsize=(12,5), grid=True,
color='red',title="Number of customers per month in red, per week in blue and per day in black").set(xlabel="Date", ylabel="Total Number of Items Sold")[Text(0.5, 0, 'Date'), Text(0, 0.5, 'Total Number of Items Sold')]
#The number of sales per customer for every day, every week, and each month:
fig3 = plt.figure("Figure 3")
ax = plt.axes()
day_ratio = Date.resample("D")['itemDescription'].count()/Date.resample('D')['Member_number'].nunique()
day_ratio.plot(figsize=(12,5), grid=True,
color='black').set(xlabel="Date", ylabel="Sale per customer")
week_ratio = Date.resample("W")['itemDescription'].count()/Date.resample('W')['Member_number'].nunique()
week_ratio.plot(figsize=(12,5), grid=True,
color='blue').set(xlabel="Date", ylabel="Sale per customer")
month_ratio = Date.resample("M")['itemDescription'].count()/Date.resample('M')['Member_number'].nunique()
month_ratio.plot(figsize=(12,5), grid=True,
color='red', title = "Sale per customers per month in red, per week in blue and per day in black").set(xlabel="Date", ylabel="Sale per customer")[Text(0.5, 0, 'Date'), Text(0, 0.5, 'Sale per customer')]
#5 best seller items
Item_distr = dataset.groupby(by = 'itemDescription').size().reset_index(name='Frequency').sort_values(by = 'Frequency',ascending=False).head(5)
bars = Item_distr["itemDescription"]
height = Item_distr["Frequency"]
x_pos = np.arange(len(bars))
plt.figure(figsize=(16,9))
plt.bar(x_pos, height, color = 'blue')
plt.title("Top 5 Sold Items")
plt.ylabel("Number of sold items")
plt.xticks(x_pos, bars)
plt.show()
The most well-known algorithm for association rule learning is the Apriori algorithm, which works by identifying frequent itemsets (i.e., combinations of items that occur together frequently) and then generating association rules from those itemsets. An association rule is a statement that indicates the likelihood of one item being purchased given that another item has been purchased.
For example, if customers who buy milk and bread also tend to buy eggs, then an association rule could be "milk and bread imply the purchase of eggs". These rules can be used to provide insights into consumer behavior, to inform marketing strategies, and to make recommendations for complementary products.
##Data preparation and modeling
## Before modeling, the transaction must be one-hot
Transactions = dataset.groupby(['Member_number', 'itemDescription'])['itemDescription'].count().unstack().fillna(0).reset_index()
Transactions.head()
def one_hot_encoder(k):
if k <= 0:
return 0
if k >= 1:
return 1
Transactions = Transactions.iloc[:, 1:Transactions.shape[1]].applymap(one_hot_encoder)
# Transactions.head()
##associate learning 1: apriori
frequent_items1 = apriori(Transactions, min_support=0.027, use_colnames=True, max_len=3).sort_values(by='support')
frequent_items1.head(10)
results1 = association_rules(frequent_items1, metric="lift", min_threshold=1).sort_values('lift', ascending=False)
results1 = results1[['antecedents', 'consequents', 'support', 'confidence', 'lift']]
results1.head(100).dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| antecedents | consequents | support | confidence | lift | |
|---|---|---|---|---|---|
| 629 | (sausage) | (yogurt, rolls/buns) | 0.035659 | 0.173101 | 1.554717 |
| 624 | (yogurt, rolls/buns) | (sausage) | 0.035659 | 0.320276 | 1.554717 |
| 216 | (root vegetables, whole milk) | (shopping bags) | 0.029246 | 0.258503 | 1.536046 |
| 221 | (shopping bags) | (root vegetables, whole milk) | 0.029246 | 0.173780 | 1.536046 |
| 627 | (yogurt) | (rolls/buns, sausage) | 0.035659 | 0.126020 | 1.530298 |
| ... | ... | ... | ... | ... | ... |
| 1008 | (rolls/buns, whole milk) | (sausage) | 0.048743 | 0.272989 | 1.325167 |
| 805 | (whole milk) | (yogurt, bottled water) | 0.040277 | 0.087906 | 1.323001 |
| 800 | (yogurt, bottled water) | (whole milk) | 0.040277 | 0.606178 | 1.323001 |
| 725 | (rolls/buns, bottled beer) | (whole milk) | 0.038225 | 0.605691 | 1.321939 |
| 728 | (whole milk) | (rolls/buns, bottled beer) | 0.038225 | 0.083427 | 1.321939 |
100 rows × 5 columns
Frequent Pattern (FP) growth is a popular algorithm used for mining frequent itemsets in a transactional dataset. The FP-growth algorithm builds a compact data structure, called a frequent pattern tree (FP-tree), to represent the transactional dataset. This data structure enables efficient counting of frequent itemsets by compressing the original dataset into a set of conditional databases. The algorithm recursively constructs the FP-tree by repeatedly finding frequent items and creating conditional databases, which are then recursively processed until no more frequent itemsets can be found.
##associate learning 2: fpgrowth
frequent_items2=fpgrowth(Transactions, min_support=0.027, use_colnames=True, max_len=3).sort_values(by='support')
frequent_items2.head(10)
results2 = association_rules(frequent_items2, metric="lift", min_threshold=1).sort_values('lift', ascending=False)
results2 = results2[['antecedents', 'consequents', 'support', 'confidence', 'lift']]
results2.head(10).dataframe tbody tr th {
vertical-align: top;
}
.dataframe thead th {
text-align: right;
}
| antecedents | consequents | support | confidence | lift | |
|---|---|---|---|---|---|
| 627 | (sausage) | (yogurt, rolls/buns) | 0.035659 | 0.173101 | 1.554717 |
| 622 | (yogurt, rolls/buns) | (sausage) | 0.035659 | 0.320276 | 1.554717 |
| 208 | (root vegetables, whole milk) | (shopping bags) | 0.029246 | 0.258503 | 1.536046 |
| 213 | (shopping bags) | (root vegetables, whole milk) | 0.029246 | 0.173780 | 1.536046 |
| 624 | (rolls/buns, sausage) | (yogurt) | 0.035659 | 0.433022 | 1.530298 |
| 625 | (yogurt) | (rolls/buns, sausage) | 0.035659 | 0.126020 | 1.530298 |
| 695 | (sausage) | (yogurt, other vegetables) | 0.037199 | 0.180573 | 1.500795 |
| 690 | (yogurt, other vegetables) | (sausage) | 0.037199 | 0.309168 | 1.500795 |
| 369 | (shopping bags) | (other vegetables, soda) | 0.031042 | 0.184451 | 1.485518 |
| 364 | (other vegetables, soda) | (shopping bags) | 0.031042 | 0.250000 | 1.485518 |