Introducing
Your new presentation assistant.
Refine, enhance, and tailor your content, source relevant images, and edit visuals quicker than ever before.
Trending searches
3
Weekly Sales Predictions
Walmart Dataset
Abhinav Dharmadhikari
Anish S Iyer
Pramod Govindarjan
Sangita Yemulwar
Rahul Rathi
Introduction
Retail Industry
Walmart
Analytics
Objective
Our motivation
Business
Problem
Weekly Sales Prediction
One challenge of modeling retail data is the need to make decisions based on limited history. Holidays and select major events come once a year, so how does the chance to see In addition, markdowns which are the known factors affect sales – the challenge is to predict Sales.
Effects of Analytics
Data
“If we have data, let’s look at data. If all we have are opinions, let’s go with mine.” — Jim Barksdale
Data
Dataset Context
Dataset
Historical sales data for 45 stores located in different regions - each store contains a number of departments.
The company also runs several promotional markdown events throughout the year. These markdowns precede prominent holidays, the four largest of which are the Super Bowl, Labor Day, Thanksgiving, and Christmas. The weeks including these holidays are weighted five times higher in the evaluation than non-holiday weeks.
Sales Dataset
Data Properties
- Dataset Shape ( 4lk , 5 )
- Granularity : agg_by(store , dept , week)
Data Distribution
Weekly Sales
1
Right Skewed
Is_Holiday
1
Ratio of holiday to non-holiday is 2 : 23
Feature Dataset
Feature Dataset
Data Properties
- Dataset Shape ( 8k , 12 )
- Granularity : agg_by(Store , Date)
Data Distribution
Markdown
1
Missing Values : max 64%
Distribution : Skewed
Features
1
Date Count : 1
Object Count : 1
Numerical Count : 10
Store Dataset
Stores Dataset
Data Properties
- Dataset Shape ( 45 , 3 )
- Gralularity : agg_by(store , type)
Data Distribution
Size
1
slightly left skewed
Features
1
Object Count : 1
Numerical Count : 2
Then
store
dataset merged
sales
&
features
Final Dataset contains all int intersection data of all the dataset
Merged All Three Dataset
Merged Dataset
Final Dataset
Data Properties
- Dataset Shape ( 4lks,16 )
- Granularity : agg_by(Store , Dept , Date)
Data Distribution
Distribution :
1
Mostly Right Skewed & Multi-Model
Missing Values : max 75%
Imputation Strategy
1
Filled with ZERO
Filled with MEDIAN
Features
1
Date Count : 1
Object Count : 4
Numerical Count : 11
EDA
Insights
Weekly Sales
Time Series Analysis
Data Prep
OLS Model Summary
1
The OLS Model calculated 8% as adj-R2 value which is not expected we test the data with some non-linear models and there we get about 90% so there we concluded that the data we have some sort of non-linear relationship
Outlier Treatment
1
In Retail Data there are no outlier , they are just extreme values.
To identify outlier we should fit a best fit line to the data and analyze the data point.
Transformations
1
Our data is influence by the time so there are time series decomposition & transformation can applied but after analysis we found that the data has inconsistency in date and about 1 year is missing.
Modeling
Approach
Hyper-Parameter
Tuning
LS
Base Models
Approach
Cleaning
Optimization & Deployments
Feature Engineering
&
Selection
BASE MODEL
INSIGHTS
Base Models
Feature Selection Insights
Feature
Selection
Summary
Best Method and Decision Function
1
Recursive Elimination with Linear Regression
Best Model
1
Decision Tree Regression with 12 Feature
Best Model
Best Model : Decision Tree
1
Summary : Over Fitted + 12 Feature
2
Selection Citation : RMSE
3
Hyper - Parameter Optimization
After Analyzing the plot we got two factors which are effecting , by using Random Search Finding best combination
Bagging & Hyper - Parameter Optimization
Using bagging RandomForest we are getting about 95% Accuracy with 6k RMSE
Conclusion
Final Commits
Business Effects
- Smarter tech to inform decision-making
- A streamlined supply chain
- Continued focus on mobile
- Customer relations
- Recruitment practices
- Revenue Optimization
Data Gathering & Quality
After working with this dataset we feel like there is some data missing and the description of the data is not properly documented. Next time we can find more data that use and can find the more relevant trends.
Also the data quality is not up to the mark. Data is inconsistent and about 75% of the data is missing,Next time we can find more data inside of imputation.
Improvements Possible
- Data Quality Could be better.
- More Feature can gathered.
- We can use clustering and cluster analysis for discovery.
- Time Series Analysis can be used which might reveal hidden pattern.
- Time Series Forecasting can be used inside of regression which might given more better results.
- the given data does not show the business cycle , Next improvement can be provided by domain expert about the business cycle.