Session #3 - Supply Chain

Key Notes

• Advances of Tech
• Adoption of Digital Tech
• Everyone has a control tower

• Improve yourself, family, country, the universe
• Supply Chain Action
• Enterprise-level - DBT
• People Centricity, Location Independence

• Transparency, Optimization, Sense demand

• Rebuild SCM

• Key Tech - IoT, RFID, Cloud, AI
• Sensors for blood plasma monitoring
• 50 times ROI
• 3D printing

• Hyperlocal - print locally, distribute faster

Summary - Interesting use case on the impact of hyperlocal manufacturing

cs229 Interesting Projects - FinTech

Project #1 - Loanliness: Predicting Loan Repayment Ability by Using Machine Learning Methods

Key Notes

• Evaluating and predicting the repayment ability of the loaners is important for the banks to minimize the risk of loan payment default

Consumer Financial Protection Bureau rules

• Expected income or assets
• Employment status
• Expected monthly payment
• Monthly payment on the simultaneous loans
• Monthly payment of the mortgage
• Current debt status
• Residual income
• Credit history

Data Pre-processing

• Feature concatenation
• Feature Encoding and Normalization
• Invalid/Empty Entry Replacement
• Polynomial feature transformation

Summary - Data pre-processing, classification algorithms lessons

Code - Link

Project #2 - Detecting Credit Card Fraud with Machine Learning

Key Notes

Class imbalance solutions

• Undersampling – balances the data by randomly choosing observations from the majority class to exclude
• Oversampling – balances the data by randomly oversampling the minority class
• Both – a hybrid method that employs both undersampling and oversampling
• ROSE – a synthetic data generation method that balances the data by creating artificial samples of the minority class in the neighborhood of existing examples

Models

• Logistic Regression with Quadratic Terms and LASSO Regularization
• Simple Logistic Regression
• Random Forests
• Neural Networks

Summary - Handling imbalanced data, Applying different ML algos

Project #3 - Algorithmic Trading using LSTM-Models for Intraday Stock Predictions

Key Notes

• Feature Extraction - min-max-scalar

Keep Thinking!!!

Supply Chain session #2 - Systems Analytics Global Leaders' Seminars: Professor Stefan Minner, Technical University of Munich

Key Notes

Data Driven approach to strategic inventory placement in multi-echelon network

• Gartner Analytics offering

• Future + Predictive Analytics
• Data Driven inventory management
• Integrated tasks (Forecasting + Optimization)
• Product shelf life, persishable goods

• Forecasting + Optimization in an integrated way

• Pricing challenges in retail
• Demand profiles

• Data driven optimization
• Setting inventory levels
• Forecasting inventory level
• Competitor prices

• Holding cost vs unsatisfied demand

• Regression
• Multivariate regression
• Linear programming
• Mixed-integer programming model

Books

Scheduling in Supply Chains Using Mixed Integer Programming

Link2, Link3, Link4

Keep Thinking!!!

Linear programming - Optimization Problem - Tools

Linear programming is a simple technique where we depict complex relationships through linear functions and then find the optimum points.

Let us define some terminologies used in Linear Programming using the above example.

Decision Variables: The decision variables are the variables which will decide my output.They represent my ultimate solution. To solve any problem, we first need to identify the decision variables.

Objective Function: It is defined as the objective of making decisions.

Constraints: The constraints are the restrictions or limitations on the decision variables.They usually limit the value of the decision variables.

PuLP – LP Solver Front-End

LpProblem - Defines the LP problem. Holds the constraints and objective function.Interface to the LP Solver (external)

LpVariable - Abstracts an LP variable (with name). Values will be changed by the solver.Float or integer. Defines the permitted variable value range

LpConstraint – Constraint rule. Can be one of: <=, =, >=

Keep Learning!!!

Supply chain sessions #1 - Systems Analytics Global Leaders' Seminars: Professor Christopher Tang, UCLA (October 7, 2020)

Key Notes

• Faster Delivery, Smart Order Assignments on On-demand meal delivery platform

• Mobile / Internet-based / They do not own content
• High value, Zero ownership of assets
• Platform types (Ecommerce / Resource sharing)
• Value creation / Markets creation

• Money made by commissions / Free for customer
• Impact and benefits - Wider impact, New jobs creation

• Scopus / Google Scholar for papers
• Meal delivery platforms
• Order meals, App, list restaurants, delivery partner

• Platform overview
• Three-sided market
• Platform = Customers, Restaurants, Drivers
• Charge customers + Restaurants
• Required time (area of research)

• Platform operations
• App overview
• Price
• Order time 
• Delivery Time Estimation
• Preorder for next day 

• Polling to schedule drivers 
• Wait time at restaurants
• On-time / Delayed Delivery
• Factors affecting delivery performance
• Future customer perspectives

• Behavior for future orders
• Shangai city data

Insights

• Distribution for Long orders by time(By Distance)
• Distribution for Short orders by time  (By Distance)
• Peak Hours
• Average delivery time for different routes
• Average delivery time for different restaurants
• %% of Ontime delivery by time
• %% of Early vs Late orders
• Repeating orders 
• Pricing by peak hour / demand / rule based / recommendation based
• Offer conversions
• Revenue metrics by Restaurants
• Revenue metrics by Days
• Revenue metrics by Customer Segments
• Driver delivery metrics 
• Congestion levels of Traffic / Temperature / Seasonality
• Driver familiarity with the route

Use cases

• Customer churn
• Driver churn
• Next order prediction based on intervals
• Repeat order customer
• Rare customer
• Empirical work before prescriptive
• Performance on future orders
• Survival analysis

Models

Aalen Additive model

• Regression model for delivery performance time with features like experience, restaurant distance, order estimated time, waiting time

 

• Delivery experience vs Future orders relationship

Estimate delivery time for order

Driver location estimation for assignment

Total delivery time for order

Total delay

Maximum future orders

Discourage late delivery with incentives 

 

Predicting Shipping Time with Machine Learning

Package Delivery Time dataset

Using Machine Learning computing, we developed a model capable of improving the predictability of shipping times

Key Notes

The important milestones for this project are, in chronological order:

• The booking date, which is the date when the transport from origin to final destination is booked (called “book date” in the dataset);
• The receive date, which is the date when the cargo is received by the shipper at the port of origin (called “Actual Receipt Date” in the dataset);
• The gate-in date, which is the date when the cargo enters the yard where it will wait to be loaded on the vessel (called “Gate In Origin-Actual” in the dataset);
• The vessel departure date, which is the date when the cargo leaves the port of origin (called “Vessel Actual Departure” in the dataset)
• The vessel arrival date, which is the date when the cargo arrives at the port of destination (called “Vessel Actual Arrival-Actual” in the dataset)
• The unload date, which is the date when the cargo is unloaded from the vessel (“Container Unload From Vessel-Actual”)

Selecting relevant external factors

• Chinese New Year
• Port Congestion US
• Weather
• Positioning of the Container on the vessel

Selecting features for the Machine Learning model

• Average time spent at Port of origin (Port of origin mean) - The average is based on the different shipper and port combinations
• Standard deviation of time spent at Port of origin (Port of origin SD) - The standard deviation is based on the different shipper and port combinations
• Average time spent at Port of destination (Port of destination mean)
• Standard deviation of time spent at Port of destination (Port of destination SD)
• Average time spent per Route (Route mean)
• Schedule
• Origin Service
• Holiday
• Quarter
• Expected Time to port
• Port of origin late
• Port of origin latest
• Capacity of port of destination
• Late departure

From startup - https://nextbillion.ai/blog

Key Metrics

• Delivery time = Promised ETA ± Threshold time for the territory
• Promise kept metric = (Number of accurate orders / total orders of the day) X 100%
• For the end-consumer, the role played by an ETA begins even before the service request is placed
• For the delivery agent/driver, an accurate ETA sets a service promise that needs to be met.
• A great metric to track the impact of ETAs on pricing is the % of orders/rides that are priced accurately.

Keep Thinking!!!