Business Knowledge Series course
Presented by Prof. Dr. Gero Szepannek, University in Stralsund
Business Knowledge Series courseThis business-focused course provides the necessary knowledge to choose between traditional techniques and state-of the art machine learning algorithms and optimize your predictive data mining models with respect to your business objectives.
Learn how to
- integrate business considerations to model selection
- tune and optimize your models
- properly compare predictive performance of models
- understand the black box resulting from modern machine learning algorithms.
Who should attend
Data scientists with a practical or research background who want to find a suitable predictive model for a given application.
Formats available | Standard duration | | |
Classroom: |
1 day | | |
|
Attendees should have experiences with the use of logistic regression and decision trees as well as basic knowledge on standard performance measures like misclassification error or AUC in the predictive modelling context, which are content of the Applied Analytics using SAS Enterprise Miner (AAEM) course. For the examples it is further helpful to be familiar with using SAS Enterprise Miner. Business experience is preferred (but not necessary), particularly in the following sectors:
- CRM
- Credit risk
- Industry 4.0
- Medical informatics.
Generally, the content of the course is software-neutral but there is emphasis on SAS Enterprise Miner implementations.
Traditional vs. Modern Algorithms- Overview on modern machine learning techniques: Trees, SVMs, Random Forests and Boosting
- What makes a good model? Bias and Variance in a decision tree example
- Relation to model parameters
- The effect of parameter tuning
- Understanding the effect of hyperparameters
- Determine the appropriate degree of model complexity
- Proper comparison of models for different performance measures
Case Study: Scorecard modelling using random forests- Potential benefits and risks of modern algorithms
- Challenging the gold standard
- Effect of preprocessing
- Business considerations
- On the randomness of random forests
- The value of expert knowledge integrating
Benchmarking and Tuning Models- Systematic model parameter optimization
- Let the computer tune your model
- Connecting SAS Enterprise Miner and R for model tuning
- Benchmarking your model
- Feature engineering
- Roadmap
- Lesson learned from experience
Opening the Black Box- Pros & cons of state of the art algorithms
- Variable importance
- Variable importance within SAS Enterprise Miner
- Partial dependence
- Possible extensions to SAS Enterprise Miner
