Business Overview

Demand Forecasting is a process by which the future customer demand of any product is estimated by using the available data for the product. Demand forecasts are important to the most basic processes in any organization. To plan and deliver products and services is necessary to know what the future might hold. A demand forecast is important to plan all business decisions: sales, finance, production management, logistics and also marketing. To be able to predict next purchases is a valuable thing for any business. If we make the best forecast using data then it can help businesses to provide consumers with the right product at the right place at the right time which is very much important to business to get more profit margin.

Aim

To Forecast Inventory demand using historical sales data of Grupo Bimbo bakery products in R.

Dataset: https://www.kaggle.com/c/grupo-bimbo-inventory-demand/data

Data Description

The dataset used is from Mexican multinational company, Grupo Bimbo. The delivery chain is present in countries of America, Europe, Asia, and Africa. It has an annual sales volume of 15 billion dollars. Grupo Bimbo delivers fresh bakery products to 1 million stores along its 45,000 routes across Mexico. There are five datasets available.

1. train.csv: It contains data available for training. The dataset contains about 7.4 billion entries over 11 features.
2. train.csv: It contains data available for testing.
3. cliente_tabla.csv: It contains the data of 93 million clients over 2 different features.
4. producto_tabla.csv : It contains the data for 2592 products.
5. town_state.csv : It contains the data for 790 towns.

The basic information about the features available in data are as follows.

• Semana - Week number (From Thursday to Wednesday)
• Agencia_ID - Sales Depot ID
• Canal_ID - Sales Channel ID
• Ruta_SAK - Route ID
• Cliente_ID - ID of client
• NombreCliente - Name of the client
• Producto_ID - ID of product
• NombreProducto - Name of the Product
• Venta_uni_hoy - Number of units sold in this week in integer
• Venta_hoy - Weekly sales in unit pesos
• Dev_uni_proxima - Number of units returned in next week in integer
• Dev_proxima - Next week returns in unit pesos
• Demanda_uni_equil - Demand adjusted in integer (Target variable)

Tech Stack

• Language : R
• Libraries : dplyr, ggplot2, caTools, xgboost, data.table, Matrix, lightgbm, gbm, caret, zoo, DataCombine, e1071

Approach

1. Exploratory data analysis (EDA)
1. Data visualization
2. Inference about features
3. Feature engineering

2. Data cleaning (outlier/missing values/categorical)
1. Missing value detection
2. Replacing special characters in data

3. Feature Engineering
1. Adding lag columns for target variable
2. Adding moving average columns for target variable

4. Model building on training data
1. XGBoost
2. GBM
3. SVM

5. Model validation
1. RMSE

6. Conclusion