Data Science & BI an der International School of Management

Data Science deals with predictive & prescriptive analysis. That means it deals with the unknown unknowns, so it calculates best predictions without given formulas.

Business intelligence deals mainly with descriptive analysis. That means it deals with known unknowns, so it calculates KPIs and predictions using known formula.

Data Scientists are more knowledgable about data & analytics.

BI professionals understand the needs & requirements to build & maintain BI solutions & reports

Descriptive analytics: use statistics to describe the data used to gain information or other purposes.

Predictive analytics: makes predictions about unknown future events, discloses reasons behind them typically by advanced analytics

Prescriptive analytics: optimizes indicators and recommends actions for smart decision-making

Big data: 

• data that are too large / too complex
• fast processing time
• many users
• cannot be effectively and/or efficiently handled by traditional data-related theories, technologies and tools

Advanced analytics:

• theories, technologies and tools to get in-depth understanding of Big data

Data analysis /analytics:

• theories, technologies and tools that enable in-depth understanding into data
• consist of descriptive, predictive and prescriptive analytics

Explicit analytics:

• Focus on descriptive analytics
• involving observable aspects by reporting, descriptive analytics, alerting & forecasting

Implicit analytics:

• focus on deep analytics
• involving hidden aspects by predictive modeling, optimization, prescriptive analytics and actionable knowledge delivery

Deep analytics:

• refers to data analytics
• in-depth understanding of why & how things have happened, are happening or will happen
• cannot be addressed by descriptive analytics

Traditional systems:

• structured data
• transaction systems
  • ERP, SCM, PLM - product data, HR data, customer data

New systems:

• semi-structured data (XML)
• unstructured data (text/ files)
• machine systems / sensor data
  • manufacturing data, weather information
• position data (mobile / IT)
  • RFID, telecom data
• Relationship & behavior data (social / semantic web)
  • twitter, linkedin, facebook

Business intelligence:

describe and understand data

(data mining, statistics)

Business analytics:

predict and optimize data

(simulation, predictive methods, optimization methods)

Both together: improve

Volume, Variety, Velocity, Value, Validity

A. Identification of management problem

1. Data preparation

2. Selection of algorithm to model data

3. Tuning model parameters for selected algorithm

4. Evaluation & Validation of models based on their accuracy 

B. Alternative solutions to the management problem

• selection of variables
• characteristic extraction
• handling of missing data (approximate, calculate, delete values)

a) unsupervised learning

k-means clustering

association analysis

principal component analysis

What patterns are hidden in my data?

We don't know which pattern we are looking for.

b) supervised learning

regression analysis

k-nearest neighbours

decision tree

derive forecasts from the patterns in my data

predictions based on predefined patterns

c) encouraging learning

A/B test, slimming epsilon

strategy (multi-armed bandits)

derive predictions from patterns in my data and improve predictions as new data arrives

improves itself by using results as feedback

Parameters:

options that allow you to vary the settings of an algorithm

Hypersensitive: 

error: overfit - can hardly be transferred to other data

Insensitive:

error: unterfit - ignores obvious structures, neglects trends, no reliable conclusions