03. Market Segmentation (W3)

Descriptive Analytics - Market Segmentation

Market segmentation

• Segmentation
  • Males and females of different shapes and sizes, children – clothing, shoes
  • Climate or cultural differences
• Targeting or Positioning
  • Target some distinctive segment
  • Position your company as belonging or appealing to some segment
• Examples
  • Peroni Beer – Italian beer
  • Italian pizza

How might researcher choose to segment?

Measurable, Durable, Accessible, Different, Substantial

Segmentation base/variable	Examples
Demographics	Age, gender, income, occupation, education, family size, geography
Psychographics	Attitudes, opinions, activities, personality, lifestyle, interests, values
Behavioural	Usage rate, main brand, media used
Other	Occasion/situation, benefits sought, media habits

Database-driven targeting

• Recency, Frequency, and Monetary value (RFM)
• Existing customers base VS new customers
• Heavy VS Light customers

Methods for segmentation

• Business rules
  • Arbitrary criteria from demographics, psychographics and/or behavior
  • Arbitrary quantile /threshold membership from above characteristics
• Clustering
  • Unsupervised methods of data analysis

Review Clustering

Unsupervised classification

1. Choose meaningful variables
2. Select a measure of a distance or similarity/dissimilarity
3. Maximize between group distance and minimize within group distance
4. Interpret resulted clusters

Selection of variables

Your variables:

• are meaningful for the analysis objective
• are (relatively) independent
• are limited in number
• are numerical (one hot encoding for categorical)
• have low kurtosis and skewness statistics (at least in the training set)

One-hot encoding

library(data.table)
library(mltools)
customers <- data.frame(
    id=c(1, 2, 3, 4),  
    gender=c('M', 'M', 'M', 'F'), 
    mood=c('happy’, 'sad', 'happy’, 'sad')
)

customers <- one_hot(as.data.table(customers))
customers

Measure of distance

• Manhattan, Euclidian, Minkowski distance: $D(X,Y)=({\displaystyle \sum _{i=1}^n|x_i-y_i{|}^p{)}^{\frac{1}{p}}}$
• Cosine distance: $similarity=cos(\theta )=\frac{A\cdot B}{\Vert A\Vert \Vert B\Vert }=\frac{\sum _{i=1}^n{A}_i{B}_i}{\sqrt{\sum _{i=1}^n{A}_i^2}\sqrt{\sum _{i=1}^n{B}_i^2}}$
• Jaccard distance: $J(A,B)=\frac{A\cap B}{A\cup B}=\frac{|A\cap B|}{|A|+|B|-|A\cap B|}$

Clustering types and algorithms

• Hierarchical clustering (AGNES, DIANA)
• Partition-based clustering (k-Means, k-Medoids)
• Mean-shift clustering
• Density-Based Spatial Clustering of Applications with Noise (DBSCAN)
• Expectation–Maximization (EM) Clustering

Example: Tourist Risk Taking

The data set contains 563 respondents who state how often they take risks from the following six categories:

1. recreational risks: e.g., rock-climbing, scuba diving
2. health risks: e.g., smoking, poor diet, high alcohol consumption
3. career risks: e.g., quitting a job without another to go to
4. financial risks: e.g., gambling, risky investments
5. safety risks: e.g., speeding
6. social risks: e.g., standing for election, publicly challenging a rule or decision

Respondents are presented with an ordinal scale consisting of five answer options [NEVER(1), RARELY(2), QUITE OFTEN(3), OFTEN(4), VERY OFTEN(5)].

Code On Google Colab

Hierarchical clustering

risk <- read_csv("risk.csv")
dim(risk)
head(risk)

colMeans(risk)

risk.dist <- dist(risk, method = "manhattan")
risk.hcl <- hclust(risk.dist, method = "complete")
risk.hcl

plot(risk.hcl, main = "", labels = FALSE)

c2 <- cutree(risk.hcl, h = 20)
table(c2)
c6 <- cutree(risk.hcl, k = 6)
table(c6)

c2.means <- aggregate(risk, list(Cluster = c2), mean)
round(c2.means[, -1], 1)
c6.means <- aggregate(risk, list(Cluster = c6), mean)
round(c6.means[, -1], 1)

DBScan

library(dbscan)
library(readr)
library("factoextra")
risk <- read_csv("risk.csv")

res.db <- dbscan(risk, eps= 1.5, minPts = 5)
res.db

fviz_cluster(res.db, risk, geom = "point")

EMCluster

library(EMCluster)

emobj <- simple.init(risk, nclass = 6)
risk.em <- emcluster(risk, emobj, assign.class = TRUE)

par(mfrow = c(1, 1))
plotem(risk.em, risk)

summary(risk.em)

Cluster Plot

library(cluster) 
clusplot(risk, risk.em$class, color=TRUE, 
         shade=TRUE, labels=2, lines=0)

em.means <- aggregate(risk, 
                      list(Cluster = risk.em$class), 
                      mean)
round(em.means[, -1], 1)

Summary

• Segmentation – to do or not to do?
• Methods for segmentation: business rules vs clustering
• Variables to consider
• Distance measures
• One-hot encoding
• Clustering algorithms
• Interpretation of clusters

Reference

Tim's slides of Week 3