One of the biggest issues face by chain stores in the retail industry is the Supply chain management. The component of the supply chain involved in determining how best to fulfil these requirements are created from the Demand Plan. The objective of demand plan is to balance supply and demand in a manner that achieves the financial and service objectives of the enterprise.
If we look into the case of a retail chain store, the basic issue is to know the demand of products that are sold in the store. If the decision making authority is aware of the demand of each product on a weekly or monthly basis, they would be able to plan the supply chain accordingly. If the creation of a demand prediction model is possible, this would save the company a lot of money by preventing over-stocking and being able to plan their logistics and staffing accordingly.
Objective
To create a model to predict the weekly sales for a chain of stores by analysing the trends of sales over time and the economic features.
Data
To start this project I am using a dataset from Kaggle.com. I would be doing my analysis and study on a dataset that gives me the weekly sales data of Walmart.
There are 3 Datasets
- Weekly Sales data of 98 departments in 45 stores, primary key is a combination of (Store, Dept, Date).
- The Features dataset with economic details like CPI, Unemployment rate, Fuel price etc.. For each week in the region of the Store primary key here is a combination of (Store, Date)
- In the third dataset I have area of each store and Type primary key is Store
Method
I am breaking this problem into 2 stages
1. Create a predictive model that predicts the weekly sales of 45
Stores (In this stage I am not planning to predict the sales on
Department Level)
2. Extend the project to Department Level (Future Work)
I did some basic cleaning up of the data I had and did some basic exploratory analysis. From the insights I got from the analysis, I further moved to creating models and fine tuning of models. I used different type of Regression Models and Times Series Analysis to proceed and solve this stage.
List of Models
1. Lasso Regression
2. Random Forest Regression
3. Support Vector Regression
4. Time Series Analysis
Conclusion
I have come up with different models that have helped me to predict the weekly sales of 45 stores with around 96.5% of accuracy for 3 months(90 Days). Out of the models I ran, I selected the Lasso Regression as my predictive Model since it gave me the high accuracy of prediction with the lowest residual (error).
Even though I have selected one model as of now I am trying to predict sales for 45 stores. I am sure the impact of each feature might be different for different stores because of many reasons and thatÂ’s why I checked how different my prediction with each store is.
Here the bigger the bubble is the higher the variation in prediction. So we can see that there are 3 stores which have more than 5% variation in prediction. In the future study I can look into these store and try to see what goes wrong and using an ensemble method of getting prediction from different models try to improve the predictions
Check this in GitHub
