JUST NEED SOME TUTORING HOW I CAN APPROACH THIS MACHINE LEARNING?  

 

Your manager, XXXXX has deployed your team to assist a client with their efforts in 

classifying image data. LumenBrite is a provider of drone surveillance and analysis software used in the shipping industry for inventory tracking and management. The drones that use their software gather massive amounts of image data every day and they need a way to efficiently scrub the image data and classify images as "useful" or "not useful" based on a set of given criteria. Images that are flagged as useful are then used by LumenBrite's clients to track inventory and measure supply chain performance across a number of performance indicators.

For a typical client company, roughly 10,000 images are collected each week, and about 25% of these images are actually useful for inventory tracking. Until now, image classification has been a manual process that takes 20 hours of a staff associate's time to complete and an additional 2 hours of a manager's time to review the classifications and finalize deliverables for the client. Assume a manager bills at twice the hourly rate as an associate.

 

 

You've already developed a predictive model using training data that has 75% recall and 85% precision rates in out-of-sample tests. A human associate's recall performance is typically 60% and their precision rate is 95%. Running the predictive model takes 5 hours of an associate's time for set-up and data pre-processing each week, plus the time it takes to manually review all images that are classified as "useful" by the algorithm to filter out as many incorrect classifications as possible (assume the associate reviews these images at the same rate as in the manual approach). Because of the higher complexity of the predictive modeling approach, the manager must spend a total of 3 hours supporting and reviewing the associate's work. 

You must make the decision on whether or not to use your predictive model for this task and then defend that decision to your manager. You want to balance two considerations when making your recommendation: (1) minimizing the standardized hours required for this entire process, and (2) providing a high-quality classification output to the client, as measured by the model performance metric(s) you deem most important in this context.

 

 

Make a decision as to whether or not you would recommend the implementation of your model. Then write a report to Jeffrey explaining your decision. Be sure to include the following:

1. Your final recommendation about whether to implement the predictive model or continue with the manual approach to image classification. You should include a justification of why you chose your proposed solution over the alternative, including key measures of the cost and performance of each approach.
2. An explanation for why accuracy is not mentioned in the performance metrics for the models. Isn't this a better way to decide whether to automate or stay with the associate?
3. If your team wants to improve the classification algorithm's performance, what types of models would you recommend for this application? Is there a way to achieve better insights using an unsupervised type of machine learning?
4. Any other considerations you feel are relevant to this decision, or additional information that would be helpful to have.
5. The quantitative analysis you performed to support your recommendation, including an estimate of the standardized hours required under each approach and any performance metric calculations that were relevant to your decision, presented in an appendix to your report.