5.3.1 Specific Rules
We conclude some specific rules of return propensity for the 5 groups from the results in Section 5.2.
1. Gender (1) Female
¾ For “Category” is one of: MP3 Player, Personal Computer, Female Clothing, Female Shoes, Cameras, Music and DVDs, or Video Games and Toys.
48
“Average Delivery Days” is important to diminish return propensity (When “Average Delivery Days” is more than (include) 3 days, return propensity increases.)
Under the short delivery days (in 2 days), “Price” will be a key factor for customer to concern return. While “Price” is less than NT $338, people tend to not return.
¾ For Category is one of: Female Bags, Watches and Clocks, Sports and Outdoors, Furnishing, Computer Expendables, Maternal and Children’s, or Peripheral.
If “Price” becomes higher, then return propensity will raise in the meantime. Particularly, when “Price” is more than NT $1,998, return propensity will be obviously higher than others.
(2) Male
¾ For “Category” is one of: Laptops, Female Bags, Sports and Outdoors, Automotive, Furnishing, Bouquet and Gift, Beauty, Cameras, Appliances, Maternal and Children’s, Adult Toys, Communication, Ornament and Luxury, or Pet Supplies.
If “Price” becomes higher, then return propensity rises in the meantime. Particularly, when “Price” is more than NT $1,776.5, return propensity will be obviously higher than others.
¾ For Category is one of: MP3 Player, Personal Computer, Female Clothing, Female Shoes, Watches and Clocks, Stationery, Cosmetic, Books and Magazines, Video Games and Toys, or Peripheral.
While “Num. of Merchandise” is more than 1615.5, the return propensity becomes much higher than retailers with less number of merchandise.
While “Num. of Merchandise” is less than 1615.5, “Accumulated Buyer” of each retailer will become an index for distinguishing return propensity. Namely, the higher the “Accumulated Buyer” is, the lower the return propensity is.
2. Total Number of Transaction in 2007 (1) Equal to or less than 12 times
¾ For “Category” is one of: MP3 Player, Female Bags, Watches and Clocks, Sports and Outdoors, Furnishing, Bouquet and Gift, Beauty, Cameras, Appliances, Books and Magazines, Maternal and Children’s, Adult Toys, Communication, Peripheral, or Ornament and Luxury.
To be brief and concise, the higher the “Price” is, the steeper the return propensity is.
For more details, orders between NT $242.5 and NT $2,890 coupled with retailers providing numerous number of merchandise (more than 1,039.5), thus “Average Delivery Days” could be a factor to distinguish whether an order will be returned. In this condition, “Average Delivery Days” is better to be controlled in 2 days, or the return propensity will become much higher.
In addition, there is also a higher return propensity for orders between NT $242.5 and NT $2,890 coupled with retailers providing moderate number of merchandise (between 135.5 and 1,039.5).
¾ For Category is one of: Female Clothing, Female Shoes, Stationery, Automotive, Music and DVDs, or Furniture and Bedding.
If “Average Delivery Days” is more than (include) 3 days, then return propensity increase. If the “Average Delivery Days” is less than 2 days, then
“Price” will be a key factor for customer to concern return. While “Price” is more than NT $988.5, the return propensity will be higher than low price orders.
(2) More than 12 times
¾ For “Category” is one of: Laptops, Female Clothing, Female Shoes, Books and Magazines, Communication, or Peripheral.
While “Price” is between NT $298 and NT $2,970, the return propensity is the highest; however, return propensity decrease when “Price”
is more than NT $2,970.
¾ For “Category” is one of: Collections and Fine Works, Automotive, Male Fashion, Bouquet and Gift, Health and Personal Care, Beauty, Music and DVDs, Female Shoes, Computer Expendables, Adult Toys, Ornament and
50
Luxury, Flash Memory, or Pet Supplies.
High-transaction frequency customers tend not to return these categories no matter how much of the merchandise cost.
¾ For “Category” is one of: MP3 Player, Personal Computer, Female Bags, Watches and Clocks, Stationery, Sports and Outdoors, Furnishing, Cameras, Cosmetic, Appliances, Maternal and Children’s, Furniture and Bedding, or Video Games and Toys.
Similarly, “Price” is the key factor to distinguish if customer will return the items or not. If “Price” is higher, then the return propensity will be on the rise simultaneously.
3. Female Merchandise
Unlike the four DTs described before, “Category” is taken away from the input predicted variables. In this situation, “Price” becomes the most important variable to build classification rules.
¾ “Price” is less than NT $383:
If Average Delivery Days is more than (include) 4 days, then the return propensity will be higher than the others (Average Delivery Days is less than 3).
¾ “Price” is between NT $383 and NT $14,469:
If “Accumulated Number of Buyers” of retailers is more than 1994.5, then the return propensity will decrease. By contrast, while “Accumulated Number of Buyers” of retailers less than 1994.5 the return propensity will be higher. Particularly, a customer orders a merchandise with price between NT $1,140 and NT $14,469 from retailers within few accumulated number of buyers (less than 96), his/her return propensity will be high as well.
¾ “Price” is more than NT $14,469:
Return propensity is low for extremely high price merchandise. The result may from customer.
5.3.2 General Rules
Merchandise dimension variables are the most applicable variables for classification. Retailer evaluation function (service dimension) variables also show some directions for distinguishing return propensity. However, variables of customer dimension (Location) are short of consistence. Cross Tabulation showed average return rate is related to “Payment”, “Delivery Approach” and “Carriage” though, nevertheless, they are not selected to distinguish return behavior of each order in classifiers.
Table 5.7 summarizes the trees we conducted and reveals some important variables for different customers groups.
1. Customer Dimension
(1) Only “Age” and “Location” are chosen to be predicted variables. Even so,
“Age” only shows up in Male-DT while “Location” lack of consistence for directing return propensity.
2. Merchandise Dimension
(2) “Category” and “Price” play the major roles in identifying return propensity for different groups of customers by being selected as decision criteria at the 1st and 2nd layer.
(3) “Category”, which dominate the first layer to separate the return propensity to high (more than 60%), medium (between 50% and 55%) and low (approximately 30%) in four customer-based DTs.
(4) If “Price” is less than NT $400, then customers tend to not return. If “Price”
becomes higher, then the return propensity will increase simultaneously.
However, when “Price” is extremely high, return propensity may become lower contrarily.
3. Service Dimension
(1) Variables of service dimension- “Payment”, “Delivery Approach” and
“Carriage” are not selected in distinguishing return behavior of each order in the five models.
(2) Variables of service dimension- Service evaluation function (include
“Average Delivery Days”, “Accumulated Number of Buyers”,
“Accumulated Browsers”, “Number of Merchandise”) also represent some directions for return propensity.
52
(3) If “Average Delivery Days” is more than (include) 3 days, then customers tend to return.
Table 5.8 Summary Information of Decision Trees
Gender Total Num. of