1 of 19

Sprint 12/6/22

​

  • Tiara Adhikari
  • Pooja Choudhary
  • Xiaofeng Liang
  • Anuar Mukhambetzhanov
  • Kavindra Sahabir
  • Jenny Samaroo
  • Anthony Vallejo
  • Matthew Van Praagh

​

​

​

​

​

  • Professor Jaramillo
  • Professor Venkatesh

​

2 of 19

  • Data-informed pricing model
    • Log-Log model
    • Semi-Log model
  • Outlier detection methods
    • 1.5 IQR Rule
    • Local Outlier Factor (LOF) Method
  • Sales-lift helping decision-making
  • Back-end coding behind the scenes

Agenda:

3 of 19

*Log-Log model

​

​

​

*Semi-Log model

Assumption: Absolute change in price

results in percent change in sales

​

Assumption: Percent change in price results in percent change in sales

Data-informed pricing model

4 of 19

Semi-log model

  • For this project, we needed to implement a transformation to better analyze the data. Previously, we had used the log-log model, which took the log value of both the price and sales volume.
  • The semi-log model is also appropriate in certain scenarios (especially for SKUs with more extreme sales values)
    • The semi-log and log-log models help us understand whether products respond to price changes in a proportionate manner or an absolute manner

​

​

​

5 of 19

Log Log vs Semi Log model transformation

6 of 19

Sales lift of Gift Sets and Non-Gift Sets

7 of 19

Outliers existence and 1.5 IQR rule

  • Interquartile Range (IQR): the range between the 1st and the 3d quartiles namely Q1 and Q3 (IQR = Q3 - Q1).
  • 1.5 IQR Rule: the data points which fall below Q1 - 1.5*IQR or above Q3 + 1.5*IQR are outliers.

* This chart on the slide is from “Essentials of Statistics for Business Economics”

Extreme values &

Outliers

Q3

Q1

8 of 19

Comparing the range

Reasonable range

Q3

Q1

9 of 19

Getting clearer slope groups

Group 1

8.79%

Group 2

25.05%

Group 3

50.1%

Group 4

12.53%

10 of 19

What is Local Outlier Factor? Why do we need this method and why is it useful?

  • The Local Outlier Factor (LOF) is an algorithm that computes a score (called local outlier factor) reflecting the degree of abnormality of the observations.

​

  • We need to use the Local Outlier Factor to detect anomalous and irrational observations that magnify and skew the true picture.

​

  • It is practical since it helps in the identification and elimination of faulty records from the dataset that will subsequently be used for various analytical purposes.

​

​

​

* The chart on the slide is from scikit-learn developers webpage:

https://scikit-learn.org/stable/auto_examples/neighbors/plot_lof_outlier_detection.html

​

11 of 19

Example of how it works: SKU A1010

Outlier 1: x = 0.69897 , y = 1.2422929

​

​

​

Total observations: 16

​

Total observations: 15

Without outliers

​

​

​

12 of 19

Example of how it works: SKU A1196

Outlier 1: x = 0.47712125, y = 1.2598327

​

​

Total observations: 28

​

Outlier 2: x = 0 , y = 1.42242568

​

​

​

​

Total observations: 26

Without outliers

​

​

​

13 of 19

Non Holiday Months One Class Classifier

14 of 19

Holiday Months One Class Classifier

15 of 19

Back-end coding

Before update:

​

  1. Just for log-log model
  2. Lengthy processes

​

After update:

​

  1. Clear code
  2. Log-log and Semi-log model
  3. Date selection flexibly
  4. Attributes selection flexibly
  5. Outliers removal functions

​

16 of 19

  • Functional Implementation
  • Well-formatted
  • Well-commented
  • Easy Debugging
  • Improved Runtime and accuracy
  • Easy Updating

​

Start and End Date Functionality

17 of 19

Example(Sales lift by Clustering):

18 of 19

Example(Filtered log log and semi-log models):

19 of 19

Recap of Topics:

​

  • Data-informed pricing model
  • Outlier detection methods
  • Sales-lift helping decision-making
  • Back-end coding behind the scenes