Académique Documents
Professionnel Documents
Culture Documents
1
Target Market Segments
Target Market:
Middle to Low Socioeconomic Groups
Customer Needs:
Affordable products that meet daily needs
http://evolvedigitalagency.com/defining-target-market-101/
To achieve strategic fit, supply chain
needs to increase supply chain surplus
2
Supply Chain Surplus = Value of final product to the customer -
3
Potential Factors Affecting Sales
Inventory changes
Credit policy Seasonality Population
CPI
change
Marketing
4
Testing the Sales Factors
Inferences from Regression Analysis:
● Store, Date, Fuel_Price, CPI, Unemployment, MarkDowns 1, 3
and 5 - most significant
● MarkDowns 2 and 4 - significant
5
Types of Forecasting
Qualitative
● Market survey
● Executive opinion
● Sales analysis
Quantitative
● Naive
● Winter’s Model
● ARIMA
6
Forecasting
7
How Forecasting Adds Value to the Company
Accurate Forecasting Increases
Supply Chain Surplus by Informing
Supply Chain Decisions Regarding:
1. Uncertainty in sales
2. Aggregate planning
3. Peaks in demand like Christmas
4. Reducing stock out rates
5. Balancing between responsiveness and
efficiency
6. Drivers of Supply Chain Performance:
Facilities, inventory, transportation,
information, sourcing and pricing
http://trinamix.com/event/solution-area-supply-chain-planning/
8
Technical Presentation
9
Dataset
● The datasets were obtained from kaggle
● Combined features.csv with train.csv to obtain a combined dataframe
● NA values were replaced with zeros for markdown and median for CPI and
Unemployment rates
● The features were deemed to be not that significant for the forecast as the
regression analysis on this processed dataset failed with a very low R-sq
value.
10
Forecast for Holt Winters Model
The train set was split to validate the HW model and a forecast for the next 36
weeks were done. The MAPE value was 1.99%
11
Forecast for ARIMA Model
● An auto.arima model
was run and a the
following forecast was
obtained with an MAPE
of 1.70%
● The seasonal
component order of
(0,1,0) with moving
average component
(1,1,1) gets the best fit
12
Department Level Comparison
● Departments were chosen because they
contain similar product families which
are needed for aggregate planning
Dept
11 ● If one supply chain served these 45
stores, they would be planning by
81 product families
Departments Dept
23 ● Once the three departments were
in Population randomly chosen, all sales data was
pooled from across the 45 stores by
week (143 weeks total) for each
Dept
44 department
Based on trend and seasonality being present, Winters Model & ARIMA Models were Chosen:
Naive Model will be used to benchmark performance 14
Naive Approach
MAPE Average
Dept 11:
Key Takeaways
10.8444
1. No response to trends &
seasonality
15
Winters Model
16
ARIMA
17
Conclusion
● By comparing the MAPE and forecasting values of the three models we can say the ARIMA model
suits well and gives the best forecast with the lowest MAPE measure
● The HOLT WINTERS model also has promising results and responds to seasonality
● At the departmental level, the HW model is more consistent as compared to the ARIMA model
● For the departmental level forecast to be more accurate, additional data mentioning what products a
department sells can be used on an ARIMAX model
18