Transcript Blackwell Electronics

Electronics Retail & eCommerce
Predicting Profitability and
Customer Preferences
Diana Amador
Data Mining Activities
• Profitability prediction
• Brand preference
Criteria: Accuracy vs. Precision
• Hit the bull’s eye
▫ If I hit the bull’s eye , I am accurate
▫ If all my shots land together, I have good precision
▫ If all my shot land together and hit the bull’s eye ,
I am accurate and precise
• It is possible to hit the bull’s eye purely by
chance…
Criteria Accuracy vs. Precision
Criteria: Accuracy vs. Precision
What questions is DM answering?
• Profitability of new potential products based on
similar existent products.
• Brand Preference
Criteria: Performance Indicators
•
•
•
•
•
•
•
•
# of Correctly Classified Instances (= #)
# of Incorrectly Classified Instances (= 0)
Correlation Coeficient (= 1)
Kappa Statistic (= 1)
Mean Absolute Error ( = 0)
Root Mean Absolute Error (= 0)
Relative Absolute Error ( = 1)
Root Relative Absolute Error (= 1)
How we predict profitability of
potential products?
• Using Similarity Analysis
▫ Calculation of Euclidean distance
▫ Application of Weighting Schemes
• Using Regression Analysis
▫ IBK nearest neighbor
▫ Support Vector Machine (SMOreg)
Performance Graph
ROOT RELATIVE SQ ERROR (%)
RELATIVE ABSOLUTE ERROR (%)
SMOreg(C=1, E=2)
ROOT MEAN SQ ERROR
IBK (K=2)
MEAN ABSOLUTE ERROR
CORRELATION COEFFICIENT
0
200
400
600
800
1000
1200
1400
1600
1800
PREDICTED PROFIT SCENARIOS
• SCENARIO 1: 63% of total Profit (Game Console
199 Sony and Laptop 176 Razer)
• SCENARIO 2: 74% of total Profit (Game Console
199, Sony Laptop 176 Razer and Tablet 187
Amazon)
• SCENARIO 3: 73% of total profit (Tablet 187
Amazon, Game Console 199, Sony Laptop 176
Razer and Tablet 186 Apple)
PREDICTED PROFIT CONCENTRATION
$600.00
$500.00
$400.00
TOTAL PROFIT (K)
$300.00
CONCENTRATION (%)
$200.00
$100.00
$0.00
SCENARIO 1
SCENARIO 2
SCENARIO 3
Recommendations
• Allocate marketing/sales resources to most
profitable products
• Create special marketing for less/ more
profitable products
• Prepare to maximize profitability using
alternative scenarios
• Prepare for trade-offs
How to predict brand preference?
• K-nearest neighbor (IBk)
• J48 Tree Classifier
Criteria: Performance Indicators
•
•
•
•
•
•
•
•
# of Correctly Classified Instances (= #)
# of Incorrectly Classified Instances (= 0)
Correlation Coeficient (= 1)
Kappa Statistic (= 1)
Mean Absolute Error ( = 0)
Root Mean Absolute Error (= 0)
Relative Absolute Error ( = 1)
Root Relative Absolute Error (= 1)
Performance Criteria:
Results
7000
Brand preference
percentage differ
by approximately
5% between
surveyed and
predicted values.
6000
5000
4000
Acer
3000
Sony
2000
1000
0
Surveyed
preference
%
Predicted
preference
Recommendations
• Run Brand Preference Predictions for
products to:
▫ Update inventory
▫ Create or modify marketing and branding
plans/campaigns
▫ Ask for vendor’s collaboration and support
• Run Affinity Analysis to leverage on
collaborative marketing
How to use Data Analytics to support
decisions?
• We can make more inferences using the survey
data (Preference factors related) that we didn’t
do in this first analysis
• We can gain leverage on factors we oversee (i.e.
shipping impact)
• Find more relationships and understand how to
leverage on them to create marketing and sales
plans
• Find affinity products or factors
Data-based Decision Making
• To reduce cost of opportunity
▫ Get the best ROI
• To assess trade-offs:
▫ Higher margin
▫ Faster growth
• To create a cross-cutting corporate culture:
▫ Linking traditionally separate or independent par
ties or interests to achieve a goal