Save 37% off PRO during our Black Friday Sale! »

How to Identify Use Cases For Machine Learning

How to Identify Use Cases For Machine Learning

730ee3af6c8be2f704bae5b17b27103d?s=128

Arnab Biswas

August 01, 2020
Tweet

Transcript

  1. Identifying Use Cases For Machine Learning Arnab Biswas arnabbiswas1@gmail.com arnabbiswas1

  2. • Identify your organization’s “X” most important strategic goals •

    What problems are you facing in achieving those goals? • Would those be good problems for ML?
  3. Sources To Identify Problems Good For ML • Manual Data

    Analysis Process • Rule based decision/prediction making using software Ref: https://www.datarevenue.com/en-blog/the-ai-machine-learning-usecase-checklist
  4. Manual Data Analysis • Enlist the processes which need your

    team to spend lot of time on data analysis • For each process • Is this process of analyzing data highly repetitive? • Do you need to scale this process to achieve our strategic goals? • Do we have relevant data for that? Note : Higher volume processes are preferred Ref: https://www.datarevenue.com/en-blog/the-ai-machine-learning-usecase-checklist
  5. Software decision-making • List all software processes that repeatedly makes

    a decision or a prediction based on a set of rules • Even excel or simple scripts should be considered as software • Are there too many rules? • Are the rules too complex? • Could a human expert make a better decision than the software? Note : Higher volume processes are preferred Ref: https://www.datarevenue.com/en-blog/the-ai-machine-learning-usecase-checklist
  6. Types of data • Tabular Data (Information from Assets, Maintenance

    Records, Tickets) • Image (Photos, Maps, Videos) • Language (Details of a ticket, Customer Chat) • Audio (Recorded Calls)
  7. Validate Use cases

  8. Evaluate Use Cases High Risk High Effort Value Low Risk

    Low Effort Big bubbles at this corner are preferred Ref: https://blog.dominodatalab.com/data-science-use-cases/
  9. Evaluation Criteria • Value • Value of the knowledge gained

    by solving the use case using ML • Effort • Order of magnitude of the problem • Risk • Is the data predictive enough? • Is the volume of data sufficient? • Is the technique time tasted?
  10. References • https://developers.google.com/machine-learning/problem-framing • https://www.datarevenue.com/en-blog/the-ai-machine-learning- usecase-checklist • https://blog.dominodatalab.com/data-science-use-cases/ • https://pudding.cool/process/pivot-continue-down/

    • https://towardsdatascience.com/implementing-data-clinics-will-save- time-and-get-more-done-e1f003e1c5ee