Market Basket Analysis 實作

Step 1 – collecting data 

# 要使用Groceries的資料必須先安裝arules套件

install.packages(arules)

library(arules)

data(Groceries)

Step 2 – exploring and preparing the data 

# Groceries資料是由9835筆資料組合，包含169種品項

summary(Groceries)

transactions as itemMatrix in sparse format with

 9835 rows (elements/itemsets/transactions) and

 169 columns (items) and a density of 0.02609146

most frequent items:

      whole milk other vegetables       rolls/buns             soda           yogurt

            2513             1903             1809             1715             1372

         (Other)

           34055

element (itemset/transaction) length distribution:

sizes

   1    2    3    4    5    6    7    8    9   10   11   12   13   14   15   16   17   18   19

2159 1643 1299 1005  855  645  545  438  350  246  182  117   78   77   55   46   29   14   14

  20   21   22   23   24   26   27   28   29   32

   9   11    4    6    1    1    1    1    3    1

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max.

  1.000   2.000   3.000   4.409   6.000  32.000

includes extended item information - examples:

       labels  level2           level1

1 frankfurter sausage meet and sausage

2     sausage sausage meet and sausage

3  liver loaf sausage meet and sausage

# Groceries的資料包含每一次客戶採購的品項，列出前5筆的購物車項目（用inspect(head(Groceries))也可）

inspect(Groceries[1:5])

  items                    

1 {citrus fruit,           

   semi-finished bread,    

   margarine,              

   ready soups}            

2 {tropical fruit,         

   yogurt,                 

   coffee}                 

3 {whole milk}             

4 {pip fruit,              

   yogurt,                 

   cream cheese ,          

   meat spreads}           

5 {other vegetables,       

   whole milk,             

   condensed milk,         

   long life bakery product}

# itemFrequency可列出每一項品項佔的比例

itemFrequency(Groceries[, 1:3])

frankfurter     sausage  liver loaf

0.058973055 0.093950178 0.005083884 

＃ 用itemFrequencyPlot繪出產品佔的比例圖，support參數是僅列出此比例的項目，如不使用則會列出所有產品品項

itemFrequencyPlot(Groceries, support = 0.1)

＃ 若使用topN參數則會列出依產品排名列出

itemFrequencyPlot(Groceries, topN = 20)

# 用image繪出前5筆交易的品項

image(Groceries[1:5])

# 任意sample 100筆交易紀錄的品項

image(sample(Groceries,100))

Step 3 – training a model on the data

# 若使用原本default值的support = 0.1, confidence = 0.8得到的結果是set of 0 rules

apriori(Groceries)

parameter specification:

 confidence minval smax arem  aval originalSupport support minlen maxlen

        0.8    0.1    1 none FALSE            TRUE     0.1      1     10

 target   ext

  rules FALSE

algorithmic control:

 filter tree heap memopt load sort verbose

    0.1 TRUE TRUE  FALSE TRUE    2    TRUE

apriori - find association rules with the apriori algorithm

version 4.21 (2004.05.09)        (c) 1996-2004   Christian Borgelt

set item appearances ...[0 item(s)] done [0.00s].

set transactions ...[169 item(s), 9835 transaction(s)] done [0.00s].

sorting and recoding items ... [8 item(s)] done [0.00s].

creating transaction tree ... done [0.00s].

checking subsets of size 1 2 done [0.00s].

writing ... [0 rule(s)] done [0.00s].

creating S4 object  ... done [0.00s].

set of 0 rules 

# 加入參數為 support = 0.006, confidence = 0.25)

groceryrules <- apriori(Groceries, parameter = list(support = 0.006, confidence = 0.25, minlen = 2))

# 產生set of 463 rules

groceryrules

set of 463 rules 

Step 4 – evaluating model performance 

summary(groceryrules)

set of 463 rules

rule length distribution (lhs + rhs):sizes

  2   3   4

150 297  16

   Min. 1st Qu.  Median    Mean 3rd Qu.    Max.

  2.000   2.000   3.000   2.711   3.000   4.000

summary of quality measures:

    support           confidence          lift      

 Min.   :0.006101   Min.   :0.2500   Min.   :0.9932 

 1st Qu.:0.007117   1st Qu.:0.2971   1st Qu.:1.6229 

 Median :0.008744   Median :0.3554   Median :1.9332 

 Mean   :0.011539   Mean   :0.3786   Mean   :2.0351 

 3rd Qu.:0.012303   3rd Qu.:0.4495   3rd Qu.:2.3565 

 Max.   :0.074835   Max.   :0.6600   Max.   :3.9565 

mining info:

      data ntransactions support confidence

 Groceries          9835   0.006       0.25

# 產生的rules看起來如下

inspect(groceryrules[1:3])

  lhs             rhs                   support confidence     lift

1 {pot plants} => {whole milk}      0.006914082  0.4000000 1.565460

2 {pasta}      => {whole milk}      0.006100661  0.4054054 1.586614

3 {herbs}      => {root vegetables} 0.007015760  0.4312500 3.956477

Step 5 – improving model performance

# 用lift的欄位排序，lift的數據是呈現一個產品和另一個產品可能被同時購買的機率

inspect(sort(groceryrules, by = "lift")[1:5])

  lhs                   rhs                      support confidence     lift

1 {herbs}            => {root vegetables}    0.007015760  0.4312500 3.956477

2 {berries}          => {whipped/sour cream} 0.009049314  0.2721713 3.796886

3 {tropical fruit,                                                         

   other vegetables,                                                       

   whole milk}       => {root vegetables}    0.007015760  0.4107143 3.768074

4 {beef,                                                                   

   other vegetables} => {root vegetables}    0.007930859  0.4020619 3.688692

5 {tropical fruit,                                                         

   other vegetables} => {pip fruit}          0.009456024  0.2634561 3.482649

# 用berries這個產品來產生一個相關聯的rules

berryrules <- subset(groceryrules, items %in% "berries”)

# 可以看出哪些和berries最常在一起被購買

inspect(berryrules)

  lhs          rhs                      support confidence     lift

1 {berries} => {whipped/sour cream} 0.009049314  0.2721713 3.796886

2 {berries} => {yogurt}             0.010574479  0.3180428 2.279848

3 {berries} => {other vegetables}   0.010269446  0.3088685 1.596280

4 {berries} => {whole milk}         0.011794611  0.3547401 1.388328

# 將rules存人竹.csv檔中

write(groceryrules, file = "groceryrules.csv",sep = ",", quote = TRUE, row.names = FALSE)

# 將rules從value轉換成data.frame格式

groceryrules_df <- as(groceryrules, "data.frame”)

str(groceryrules_df)

'data.frame':     463 obs. of  4 variables:

 $ rules     : Factor w/ 463 levels "{baking powder} => {other vegetables}",..: 237 204 128 127 129 238 317 21 89 90 ...

 $ support   : num  0.00691 0.0061 0.00702 0.00773 0.00773 ...

 $ confidence: num  0.4 0.405 0.431 0.475 0.475 ...

 $ lift      : num  1.57 1.59 3.96 2.45 1.86 …