Demystifying Data Science Webinar - February 14, 2018
Download as PPTX, PDF•1 like•120 views
The document outlines a presentation by David Fussichen on data science, covering its definition, the process of machine learning, and real-world applications across various sectors such as finance, healthcare, and transportation. It provides a step-by-step approach to data science projects, including defining objectives, acquiring data, modeling, and implementing insights, with an emphasis on business applications and operationalizing analytics. Strategies for getting started in data science are also discussed, highlighting the importance of preparing data, aligning projects with business goals, and utilizing appropriate tools and skills.
![Back to the Process – Step 3: Modeling
A “model” is just a math function – the same as we learned about in algebra class
Algebra Class Model Machine Learning Model
f(x) = y f(model number, type, cycles, . . .) = y
Algebra Class Example Machine Learning Example
f(x) = 2x+3 f([features]) = [Tensor Flow Neural Network]
f(5) =13 f([specific features]) = 0.035
The modeling process uses training data to pick the most appropriate model and
determine the parameters for the model.](https://image.slidesharecdn.com/demystifyingdatasciencewebinar22018-180217222709/85/Demystifying-Data-Science-Webinar-February-14-2018-12-320.jpg)
Demystifying Data Science Webinar - February 14, 2018
- 2. Our Speaker David Fussichen Analytics8 President
- 3. AGENDA 1 2 3 What actually happens on data science projects Data Science vs BI How to get started
- 4. Data Science = Advanced Analytics What Is Data Science? DATA SCIENCE: a general term that includes the process of obtaining, transforming, analyzing, and communicating data to answer a question. MACHINE LEARNING: using algorithms and computers to do a task better as more data is fed to the algorithm. A primary type of data science analysis and a subset of AI.
- 5. The 2 Domains of Machine Learning (Unsupervised Learning) Purpose: Finding Groups (Supervised Learning) Purpose: Predictions
- 6. Real Use Cases • Fraud detection • Credit and pre-payment risk • Likelihood of default • Automatically rebalance portfolios FINANCIAL SERVICES HEALTHCARE • Predict and prevent readmissions • ICU utilization prediction • Predict payer fraud • Likelihood of adhering to regimes INSURANCE • Predict how much claim will cost • Likelihood of converting a quote to a policy • Predict fraudulent claims GOVERNMENT • Predict incoming threats • Proactively block misuse of data • Predict Medicare fraud TRANSPORTATION • Predict wear and tear • Route optimization and schedule planning MANUFACTURING, RETAIL, MAKRETING, CYBER SECURITY, CUSTOMER SERVICE, & MORE!
- 7. Step 1: Define Project Objectives Infigen Energy seeks to predict the likelihood of failure of the main bearing on each of its 1500 wind turbines. Predicting the failure of a main bearing will save up to $300,000 per year. The results of the model will be used to adjust the turbine speed to extend the life of a bearing while a replacement is planned. The request: use their data to predict turbine failure in the next 30 days.
- 8. Step 2: Acquire and Explore Data Training Data Turbine Farm Location Make-Model Cycles (M) Last Maint. Date Output 30 Day Mait. Types Events Failed 456 Forsayth 2-26 GE-Alstom-3MW ECO 100 4.7 2/5/2015 400 MW ST AR N 324 Walkaway 2 14-16 GE-Alstom -4.8MW-158 1.9 8/15/2015 375 MW ST GTR N 293 Cherry Tree 8-25 GE-Alstom -3MW ECO 100 5.8 7/13/2016 427 MW OH D Y 323 Woakwine 12-5 GE-Alstom -4.8MW-158 8.6 12/1/2017 275 MW ST NM N
- 9. Step 2: Acquire and Explore Data Training Data Turbine Farm Location Make-Model Cycles (M) Last Maint. Date Output 30 Day Mait. Types Events Failed 456 Forsayth 2-26 GE-Alstom-3MW ECO 100 4.7 2/5/2015 400 MW ST AR N 324 Walkaway 2 14-16 GE-Alstom -4.8MW-158 1.9 8/15/2015 375 MW ST GTR N 293 Cherry Tree 8-25 GE-Alstom -3MW ECO 100 5.8 7/13/2016 427 MW OH D Y 323 Woakwine 12-5 GE-Alstom -4.8MW-158 8.6 12/1/2017 275 MW ST NM N FEATURES TARGET
- 10. Skipping Ahead Turbine Farm Location Make-Model Cycles (M) Last Maint. Date Output 30 Day Mait. Types Events Failed 845 Woodline 7-23 GE-Alstom-3MW ECO 100 5.9 5/14/2016 435 MW OH NM ? 466 Lake Bonney 1-19 GE-Alstom -4.8MW-158 6.9 1/22/2014 300 MW ST GTR ? 852 Walkaway 2 6-9 GE-Alstom -4.8MW-158 11.8 9/22/2015 510 MW OH GTR ? 156 Bodangora 14-3 GE-Alstom -4.8MW-158 7.6 6/7/2017 274 MW ST AR ? Production Data before we run our model
- 11. Skipping Ahead Turbine Farm Location Make-Model Cycles (M) Last Maint. Date Output 30 Day Mait. Types Events Failed 845 Woodline 7-23 GE-Alstom-3MW ECO 100 5.9 5/14/2016 435 MW OH NM .051 466 Lake Bonney 1-19 GE-Alstom -4.8MW-158 6.9 1/22/2014 300 MW ST GTR .062 852 Walkaway 2 6-9 GE-Alstom -4.8MW-158 11.8 9/22/2015 510 MW OH GTR .371 156 Bodangora 14-3 GE-Alstom -4.8MW-158 7.6 6/7/2017 274 MW ST AR .013 Production Data after we run our model Now we can make informed decisions about turbine speed and preventative maintenance!
- 12. Back to the Process – Step 3: Modeling A “model” is just a math function – the same as we learned about in algebra class Algebra Class Model Machine Learning Model f(x) = y f(model number, type, cycles, . . .) = y Algebra Class Example Machine Learning Example f(x) = 2x+3 f([features]) = [Tensor Flow Neural Network] f(5) =13 f([specific features]) = 0.035 The modeling process uses training data to pick the most appropriate model and determine the parameters for the model.
- 13. Step 3: Modeling Modeling can be difficult to do without extensive academic training. Software helps. Tweaking a model will only result in small incremental improvements to prediction accuracy. Feature Engineering is where dramatic improvements are possible. Create new features from your existing ones to improve model performance
- 14. Step 4: Interpret and Communicate Let your BI system help you tell the story. Load model results into your system for deeper analysis. Turbine Farm Make-Model Failed 845 Woodline GE-Alstom-3MW ECO 100 .051 466 Lake Bonney GE-Alstom -4.8MW-158 .062 852 Walkaway 2 GE-Alstom -4.8MW-158 .371 156 Bodangora GE-Alstom -4.8MW-158 .013
- 16. BI Data Science Data Science vs. Business Intelligence “Operationalized Data Science” Augmented Analytics
- 17. Business Intelligence Data Science Intent Data accessibility, discovery Finding patterns in data Scope As-was / As-is Future Audience Broad Limited Output Dashboards, Reports PowerPoint Approach Dimensional Algorithmical Technology Relational, Associative Machine Learning Developers COE, Biz IT, Analysts Data Scientists BI vs. Data Science Today
- 18. Getting Started • Learning more about Data Science • The first projects • Operationalized Data Science (“Augmented Analytics”)
- 19. Learning More About Data Science STATS TOOLSCODE IT BLACK BOX ENTERPRISE SOFTWARE
- 20. • Amazon Machine Learning Tutorial • Microsoft Azure Machine Learning Studio • DataRobot Webinars • Wikipedia We’ll send these out after the webinar! Resources
- 21. Assess and prepare your data - Most companies’ data is a mess – it needs to be prepared for machine learning models. Getting Started – The First Projects Ask yourself: What does it mean to me to make this prediction? Identify business reasons for implementing data science - What value does it provide and how can it be measured? - Build models around your business objectives. Assess staffing needs. Do you have the staff, skills, and resources needed? Make a plan: complete project prioritization matrix
- 22. Getting Started - Tips 2 1 3 4 Stick with simple models Explore more problems Learn from a sample of data Focus on automation Focus on reducing the time between the data acquisition and the development of the first simple predictive model. Instead of exploring 1 business problem with incredible sophistication, explore many, and build a simple predictive model for each one. Instead of using massive computing resources to process big data, explore data subsamples which will enable the exploration of more hypotheses. Streamline manual processes, and develop algorithms to help automate. Via Harvard Business Review: “Why You’re Not Getting Value from Your Data
- 23. Moving to Operationalized Data Science • Use your corporate data • Store data in formats for human users AND machines • Use Machine Learning models to prioritize data collection efforts • Hire and use the PhDs to oversight and program design work • Train your data-literate professionals on Machine Learning • Use Machine Learning tools instead of hacking away at code • Use enterprise-class ETL tools to wrangle and organize data • Use enterprise-class BI tools to prepare training data • Use enterprise-class BI tools to tell the story So how do you get there?
- 24. Questions?