Business Analytics Q

Advanced Business Analytics

Data Mining: Arti�cial Neural Networks

Advanced Business Analytics– Majid Karimi

Neural Network Concepts

• Neural Networks (NN): a brain metaphor for information processing • Arti�cial Neural Network (ANN): computing systems inspired by the biological NNs • ANN are one of the most versatile data mining techniques that are used for:

• pattern recognition, prediction, and classi�cation

Data Mining: Arti�cial Neural Networks 2 – 19

Advanced Business Analytics– Majid Karimi

Biological Neural Networks

Data Mining: Arti�cial Neural Networks 3 – 19

Advanced Business Analytics– Majid Karimi

Processing Information in ANN

A single neuron (processing element – PE) with inputs and outputs

Data Mining: Arti�cial Neural Networks 4 – 19

Advanced Business Analytics– Majid Karimi

How is ANN inspired by NN?

Here is a great visual explanation of the the connection between ANN and NNs

Data Mining: Arti�cial Neural Networks 5 – 19

Advanced Business Analytics– Majid Karimi

How is ANN inspired by NN?

Here is a great visual explanation of the the connection between ANN and NNs

Data Mining: Arti�cial Neural Networks 5 – 19

Advanced Business Analytics– Majid Karimi

Biology Analogy

Biological ANN Soma Node

Dendrites Input Axon Output

Synapse Weight Slow Fast

Many neurons (10000000000) Few neurons (∼100)

Data Mining: Arti�cial Neural Networks 6 – 19

Advanced Business Analytics– Majid Karimi

Elements of ANN

• Processing element (PE) • Network architecture

• Hidden layers • Parallel processing

• Network information processing • Inputs • Outputs • Connection weights • Summation function

Data Mining: Arti�cial Neural Networks 7 – 19

Advanced Business Analytics– Majid Karimi

Elements of ANN: Visualized

Data Mining: Arti�cial Neural Networks 8 – 19

Advanced Business Analytics– Majid Karimi

Elements of ANN: Visualized

Data Mining: Arti�cial Neural Networks 9 – 19

Advanced Business Analytics– Majid Karimi

Elements of ANN: Transfer Functions

Linear function f(x) = x

Sigmoid function f(x) = 11+e−x

Tangent Hyperbolic function f(x) = tanh(x)

Data Mining: Arti�cial Neural Networks 10 – 19

Advanced Business Analytics– Majid Karimi

Elements of ANN: Transfer Functions (Example) Sigmoid function for Classi�cation

Data Mining: Arti�cial Neural Networks 11 – 19

Advanced Business Analytics– Majid Karimi

ANN Learning Procedure • Assign weights and calculating the output for the historical data • Calculate the errors for each observation

• Use Optimization to �nd the best weight in order to minimize the total error.

Data Mining: Arti�cial Neural Networks 12 – 19

Advanced Business Analytics– Majid Karimi

ANN Learning Procedure • Assign weights and calculating the output for the historical data • Calculate the errors for each observation

• Use Optimization to �nd the best weight in order to minimize the total error.

Data Mining: Arti�cial Neural Networks 12 – 19

Advanced Business Analytics– Majid Karimi

ANN Example: Sales Forecast

Sales Forecast

Use the data �le "Vintage.xlsx" from the “Predictive Data Mining: Arti�cial Neural Networks” folder for this example. Set up an ANN with:

• four inputs • two hidden layers of size three and two respectively. • one output.

Do not use a transformation function for this example. Use the ANN for forecasting the sales for next period of time, using the previous four historical sales.

Data Mining: Arti�cial Neural Networks 13 – 19

Advanced Business Analytics– Majid Karimi

ANN Example: Sales Forecast Continued

With the speci�cation given, the ANN structure is the following.

Input #1

Input #2

Input #3

Input #4

Output

Hidden layer 1

Hidden layer 2

Input layer

Output layer

Data Mining: Arti�cial Neural Networks 14 – 19

Advanced Business Analytics– Majid Karimi

ANN Example: Sales Forecast Continued Excel Implementation: Weight and output

Data Mining: Arti�cial Neural Networks 15 – 19

Advanced Business Analytics– Majid Karimi

ANN Example: Sales Forecast Continued Error Calculation:

Data Mining: Arti�cial Neural Networks 16 – 19

Advanced Business Analytics– Majid Karimi

ANN Example: Sales Forecast Continued Error Calculation:

Data Mining: Arti�cial Neural Networks 16 – 19

Advanced Business Analytics– Majid Karimi

ANN Example: Sales Forecast Continued Optimization and best weight calculation:

Data Mining: Arti�cial Neural Networks 17 – 19

Advanced Business Analytics– Majid Karimi

ANN Example: Sales Forecast Continued

• What are the optimal set of weights? • What is the total Mean Squared Error? • Can this ANN be further simpli�ed with less neurons?

Data Mining: Arti�cial Neural Networks 18 – 19

Advanced Business Analytics– Majid Karimi

Illuminating the “Mysterious Box" of ANN

Consider a binary classi�cation task, in which we collect data on two di�erent variables. A linear predictive model results in dividing the space in to two parts.

A Great Video to Understand the Intuition Behind Neural Networks!

Data Mining: Arti�cial Neural Networks 19 – 19