Professional Documents
Culture Documents
presentation
presentation
Finding Results
Berkay Ertürk
Contents
Objective of the project
Data sources
Data preparation
Clustering choice
Continuing the data preparation and identifying top sale increases during promotions
A correlation matrix study before building the model
Final preparation before building the model
Model choice
Model development
Interpretation of model results (promotion impact on fast/slow items & stores)
Additional model 1 (using ProductGroup1 from the Assignment4_1c.xlsx)
Interpretation of the additional model 1
Additional model 2 (using ProductGroup2 as a subset of ProductGroup1 from the
Assignment4_1c.xlsx)
Interpretation of the additional model 2
Significant difference in item return rates (bonus part)
Validation on new data
The comparison between actual and predicted sales
Recommendations for improving model results
Recommendations to marketing department for future campaigns
Conclusion
Objective of the project
Objective: Measure the effects of promotions on sales and provide
recommendations for future marketing campaigns.
Language: Python.
Outcome:
2. To calculate the average weekly sales, a new column for the week number was added.
3. Negative sales (returns) were separated for later analysis. The analysis continued with
positive sales data.
4. For better presentation, negative sales were converted to their absolute values.
6. The average weekly positive sales amounts for products and stores were calculated.
7. The average weekly positive sales amounts for products and stores were displayed in the console.
Clustering choice
Stores
Products
After using k-means clustering
Stores Products
Continuing the data preparation
and identifying top sale increases during
promotions
8. The date ranges for the first four promotions were determined.
10. The average sales for promotion and non-promotion periods were calculated.
11. The sales data for promotion and non-promotion periods were merged using inner join.
It was desired to determine if the promotion duration had any effect on the promotion
impact.
The data includes only days with promotion durations of 7 and 8. Therefore, a
significant effect was not anticipated; however, it was decided to investigate it
nonetheless.
The correlation between Promotion_Effect and Promo_Duration is 0.008848, indicating that the
promotion duration has little to no effect on the promotion impact.
18. The correlation was examined by grouping according to product categories.
Conclusion: These values indicate that the relationship between Promotion_Effect and
Promo_Duration is very weak and not statistically significant across all categories.
Final preparation before building the model
Assignment4_1c dataset were added to the combined_sales dataframe
20. The ProductGroup1 and ProductGroup2 data from the assignment4_1c dataset were added to the
combined_sales dataframe.
Model choice
Why Ordinary Least Squares(OLS) Regression?
Computational efficiency
Model development
21. Before starting model development, the data types of the columns in the
combined_sales table were examined.
Conclusion:
Conclusion:
Conclusion:
29. The negative sales amounts (returned products) during the promotion periods were filtered.
30. The negative sales amounts during non-promotion periods were filtered.
31. The return amounts from promotion and non-promotion periods were combined.
32. The return amounts for both promotion and non-promotion periods were calculated for products.
33. The return rates for both promotion and non-promotion periods were calculated for products.
34. The statistical significance of the difference in return rates between promotion and non-promotion
periods was tested using a paired t-test.
Conclusion: A statistically significant difference in return rates between promotion and non-promotion periods (p < 0.05).
The negative t-value indicates that return rates during promotion periods are lower than those during non-promotion
periods.
These results show that promotions not only increase sales but also reduce return rates.
Conclusion: In the previous analysis, it was found that promotions reduce the overall return
rate. Additionally, the impact of promotions on return rates varies by product segment.
37. The assignment4_1.b dataset was processed in the same manner as assignment4_1.a
(week number added, k-means clustering used, promotion duration included).
38. The only difference is that in this case, promo5 was used as the promotion date.
39. The new data was validated using the model developed in Additional Model 1.
The comparison between actual and predicted
sales
Conclusion:
Recommendation: The promotion duration of 7 or 8 days is not significant for the promotion effect. Therefore, we do not need to be
concerned about having a promotion duration of 7 or 8 days for future promotion campaigns.
Recommendation: Medium category products are the most positively affected by promotions. Allocate a significant portion of the
promotion budget to these products to maximize sales. Organize special campaigns and discounts for these products.
Recommendation: Slow category products are negatively affected by promotions. Develop alternative strategies for these products, such as
promotional campaigns that highlight the value and features of the products.
ProductGroup1_A Products:
Recommendation: This product group is the most positively affected by promotions. Create special and attractive promotions for these
products to significantly boost sales. Consider seasonal campaigns and bundle offers.
ProductGroup1_C Products:
Recommendation: This product group is negatively affected by promotions. Review promotion strategies and develop alternative marketing
approaches. Campaigns supported by content highlighting product features and benefits may be effective.
Category_Store_Fast Stores:
Recommendation: Fast category stores are the most positively affected by promotions. Increase the number and variety of promotion
campaigns in these stores. Special events and campaigns in these stores can attract more customers.
Recommendation: This products are the most negatively affected by promotions. Develop different marketing strategies for these products,
such as campaigns that emphasize quality and uniqueness rather than price. Work on improving the product's perception through customer
feedback and product reviews.
Recommendations
To marketing department for future campaigns
Methods: Data loading and preprocessing, data clustering, regression analysis, model evaluation.
Case Study
Finding Results
Berkay Ertürk