Introduction
- Core concepts
- Using technology effectively
- Big Data versus fat data for analytics
The Four Basic Project Types
- Predicting a value given specific conditions
- Identifying a category given specific conditions
- Predicting the next step in a sequence
- Identifying groups
The Motivation for Analytic Modeling
- Enhanced performance: An incremental strategy
- You will never have a perfect model
- Think of business as a game
- Performance metrics: Your compass to progress
- A ‘Rear View Window' perspective makes it hard to drive forward
- Conditional decision-making given expected circumstances
- The critical combination: Information & Strategy
Mathematical Modeling
- Formulas and their parts
- Anticipating outcomes from environmental conditions
- Projecting profit
- How you think about an outcome is essential
- People are inconsistent, unreliable and messy
- There is never enough data
- Samples and populations
Model Development in Three Steps
- Training
- Testing
- Validation
- A Basic Guide to the Process
Plan
- What is the business goal of the analysis project?
- What are the performance metrics for evaluating success of the decision process?
- What is the scarce resource subject to allocation?
- What is the behavior that impacts performance?
- Is there sufficient data for the target behavior to develop an adequate model?
- What is the current baseline level of performance?
- Who is on the project team?
Prepare
- What data should I include in my data sandbox?
- What does a record look like?
- What does the outcome or target variable look like?
- What data representations should I use?
- What data transformations should I use?
- How do I select variables for my model?
- How do I build my Training/Test/Validation data sets?
Build
- Algorithms give us formulas, not answers
- Formulas create a composite perspective
- There is no such thing as a 'good' algorithm
- Selecting the right tools for the job
- The environment is also not consistent
- Does the story make sense?
- Not all data is created equal
- Some data is not 'math compatible'
- Data Attributes
- Qualitative
- Nominal
- Ordinal
- Interval
- Multiple models are usually needed
- Adoption by domain experts, end users and leadership
- Project failure is not the worst outcome
- Some 'cool' features are just 'nice junk'
- Variability: Sometimes we want it, sometimes we don't
- Perfect correlation is not a good thing
- No correlation is a waste of time
Confirm
- Does our math make business sense?
- Evaluation only makes sense 'in context'
- 'Business Performance' is the only priority
- Consistent implementation strategy is critical
- Our models are in competition with each other
- How to pick a 'Challenger'
- Confirming we picked a good Challenger
Adopt
- Evaluating the expected performance of our Challenger
- Adoption by domain experts
- Adoption by end users
- Adoption by leadership
- Project failure is not our worst outcome
Replace
- Adapting to a changing environment
- The environment always changes
- Our business goals also change
- Sometimes, we just want a better answer
- Development Process by Project Type
Predicting a Value Given Specific Conditions
- Relationships in Data
- Estimating the future value of an outcome based on known current conditions
- Additional precision is more difficult to obtain and may put your project at risk
- Build (observe or work-along)
Identifying a Category Given Specific Conditions
- Shifting our thought process when the target outcome can take on a limited set of values
- The business world is not normally distributed
- The unknown attribute we are trying to predict is critical
- The misuse of regression is dangerous to your financial health – and to your business.
- Playing the odds
- The world is round and other non-linear realities
- We need a different kind of formula... sort of
- Classification is concerned with proportions, not precision
- Build (observe or work-along)
Predicting the Next Step in a Sequence
- Time series problems
- Estimating the future value of an outcome by considering the direction and distance of
- change relative to our known position
- Signal versus Noise
- Plan
- Build (observe or work-along)
Identifying Groups
- Big Data versus Big Data Analytics – The business issues
- Putting Fat Data on a Diet
- Plan
- Build (observe or work-along)
Wrap-up
- The complexity of large-scale analytics
- Specialization in project teams
- The power of adapting core analysis skills
- Where to go from here
- Predictive analytics, business intelligence and other advanced technologies
- Resources to get you on your way
Ö. A. - İş Zekası Uygulama ve Geliştirme Yönetmeni Analist
Alternatif Bank
Teşekkür ederiz.
S. K. - İş zekası geliştirme yönetmeni
Alternatif Bank
Teşekkürler
N. Q. - Uzman
Azersun Holding
Egitimden cok memnun kaldim.
G. A. - Baş Mütəxəssis
Azersun Holding
Egitimden dolayi tesekkur ederim
G. N. - CIO Assistant
Azersun Holding
Egitimden ve egitmenden cok memnun kaldim
V. H. - Süreç geliştirme uzmanı
Azersun Holding
Her şey mükemmeldi, Teşekkürler!!!
E. R. - Süreç Geliştirme uzmanı
Azersun Holding
Bir şikayetim yok, eğitici ve eğlenceli bir eğitim için teşekkür ederim
E. A. - Proje yöneticisi
Azersun Holding
İşlerinizde başarılar, Teşekkür ederim.
İ. H. - PP Consultant
Azersun Holding
Super idi! Tam iştirak edə bilməsəm də, möhtəşəm idi…