Order Status Prediction Application

About

Boss Leathers is a small leather products business which has recently started selling its products on Amazon. Currently, it has around 40 SKUs registered in the Indian Marketplace. Over the past few months, it has incurred some loss due to return orders. Now, Boss Leather seeks help to predict the liklihood of a new order being rejected. This would help them in taking necessary actions and subsequently in reducing the loss.

Objective

To build an app which would predict the order status upon providing some order details.

Data Dictionary

Order details have been scrapped from the Seller Central website using Google App - Web Scraper.

The Order data is stored in an excel file. The columns are:

Independent Features:

Column	Description
order_no	Unique Amazon Order Number
order_date	Date on which the order was placed
buyer	Name of the buyer
ship_city	Delivery Address City
ship_state	Delivery Address State
sku	Unique identifier of a product
description	Product description
quantity	Content Cell
item_total	Content Cell
shipping_fee	Charges borne by the seller to ship the item

Label / Target Feature:

• order_status - Status of the order

Tasks list:

The project is in progress and the below tasks will be completed in phases

• Order Data Analysis
• Model Building (Choosing evaluation metrics, Feature Engineering, Comparing different models)
• Create an Application
• Deploy the Application

Technologies / Tools:

• For Analysis

  • Plotly Express
  • Pandas
  • Seaborn
  • Python
  • Jupyter Notebook

• For Modelling

  • Sklearn
  • Python
  • Jupyter Notebook

• App building

  • Python
  • Flask
  • Pycharm
  • HTML
  • CSS
  • Bootstrap

• Deployment - To be decided

Future Scope

Currently, the data extraction is done manually; weekly or fornightly. Going forward perhaps this task can be automated with BeautifulSoup. Also, instead of combining multiple order details csv files, the weekly data can be appended in a common database table which can then be read for model building. To improve the accuracy we can try remodelling with different algorithms. From the frontend perspective, validations on the fields will also be added.