jnadatathon2022 / datathon

🥇 Winner project for the Accenture Challenge in Datathon FME 2022

Challenge

We were tasked to understand supply chain disruptions in the form of late deliveries and to train a model product delays based on historical inbound/outbound orders of the company. In practical terms, the task is to provide supply chain resilience in the form of disaster prevention and damage control, as well as value by reevaluating business partnerships in the iterative process of supply chain optimization.

What is it expected from the technical challenge?

• Provide data insights based on descriptive analytics of historical data
• Regression model to predict likelihood of order delay

Evaluation Criteria

What will be measured?

• Business case presentation outlining descriptive insights of main drivers’ toward order delays. How those insights can be translated to business actions and value proposition.

• Predictive model to get likelihood of order delay. ROC Curve (AUC) evaluation metric for given test data set. A Kaggle Competition have been created for teams to submit and test their model results. Please sign-up to the competition and follow the instructions

Data

Some information about the data given in the Kaggle competition.

orders.csv

Transactional historical data of the company supply chain inbound/outbound shipments

• order_id (string): unique identifier of transactional order from port inbound to final destination. Primary key of data set.
• origin_port (string): location of port where order imports arrives.
• 3pl (string): Third-party logistic company id used for distribution, warehousing, and fulfillment services.
• customs_procedure (string): Type of procedure to be used in the imports legal process
• logistic_hub (string): city name of company logistic hub address. Intermediate step between origin_port and customer
• customer (string): city name of customer destination address
• product_id (string): unique identifier of final product
• units (integer): order size quantity
• late_order (boolean): target variable, if 1 the order_id have been tagged as a late delivery, 0 is on-time

product_attributes.csv

Master data of product unit weight

• product_id (string): unique identifier of final product
• weight (integer): product weight per 1 unit in grams
• material_handling (integer): Classification id for product safety risk and risk of damage e.g. fragile, toxic, flammable.

cities_data.csv

Geographic coordinates of cities involve in the supply chain. Including distance between pair of cities

• city_from_name (string): City of location starting point
• city_to_name (string): City of destination location
• city_from_coord (tuple): Coordinates in (latitude, longitude) of city_from
• city_to_coord (tuple): Coordinates in (latitude, longitude) of city_to
• distance (float): kilometers between the pair cities

test.csv

Same as orders.csv but variable late_order has been truncated. This is the target variable

Other parts of our project

• Model inference
• Public dashboard