Neural Networks And Its Applications

1
Neural Networks AndIts Applications

• By
• Dr. Surya Chitra

2
OUTLINE

• Introduction Software
• Basic Neural Network Processing
• Software Exercise Problem/Project
• Complementary Technologies
• Genetic Algorithms
• Fuzzy Logic
• Examples of Applications
• Manufacturing
• RD
• Sales Marketing
• Financial

3
Neural Network Applications
TECHNIQUE NAME APPLICATION
Signal Processing Sensor Data Analysis, Acoustic Failure Diagnosis, Fault Detection, Speech Recognition, Noise Cancellation, Radar Processing.
Image Processing Machine Vision, Fingerprint Identification, Medical Image Processing, Industrial Inspection
4
Neural Network Applications (contd..)
TECHNIQUE NAME APPLICATION
Non-Linear Modeling Risk Analysis, Medical Diagnosis, Chemical Modeling, Real Estate Appraisal, Capacity Planning, Insurance Claim Validation.
Time-series analysis Economic Forecasting, Stock Commodity Trading, Weather Prediction, Sales Forecasting.
5
Neural Network Applications (contd..)
TECHNIQUE NAME APPLICATION
Process/ Manufacturing Management Manufacturing Process Control, Process/Product Optimization, Robotic Process Control, Inventory Management, Product Quality Control
Classification Loan Evaluation, Optical Character Recognition, Biological Cell Identification, Chemical Spectral Analysis, Customer Target Marketing, Fraud Detection.
6
NN Applications in Manufacturing
Catalyst Manufacturing Improvement
OBJECTIVE Minimize Precipitation Without
Sacrificing Quality

• 4 Inputs
• 3 Ingredients
• Solvent
• 2 Outputs
• Polymer Gel Time
• Amount of Precipitation

7
NN Applications in Manufacturing
Catalyst Manufacturing Improvement
NN TRAINING Gel Time and Amount of Precipitation
Data for Several Production Lots was used.
RESULTS Using NN Model, Operating Conditions for
Minimum Precipitation was Obtained. These
Conditions were Verified with Experiments Before
the Production Changes
8
NN Applications in RD
Development of Hydrogenation Kinetics
OBJECTIVE Develop Kinetic Model to Increase
Throughput for Hydrogenation Process

• 3 Inputs
• Pressure
• Temperature
• Catalyst Load
• One Output
• Rate Constant

9
NN Applications in RD
Development of Hydrogenation Kinetics Experimental
Plan
10
NN Applications in RD
Development of Hydrogenation Kinetics Comparison
of NN and Statistical Models
11
NN Applications in RD
Development of Hydrogenation Kinetics
RESULTS Using NN Model, Operating Conditions for
Hydrogenation Process to give Higher Reaction
Rate and Thereby Higher throughput. These
Conditions were Verified with Experiments Before
the Production Changes.
12
NN Applications in Marketing

• Modeling Customer Behavior
• Prospect Scoring
• Retention Loyalty Studies
• Profitability Analysis
• Credit Scoring
• Delinquency Behavior Scoring
• Database Enhancement
• Patchy Database interpretation

13
NN Applications in Marketing (contd..)

• Customer Segmentation
• Classify Customer-base (Unsupervised)
• Database Enhancement
• Retail Modeling
• Geo-demographic Classification
• Small-area Modeling

14
NN Applications in Marketing (Contd..)

• Sales Analysis
• Multivariable Sales Data
• Advertising promotions
• Competition macro-economics
• Alternate to Time Series Forecasting
• Data Visualization
• Distill Highly Noisy Data
• Graphically Present Clean Data

15
NN Applications in MarketingProspect Scoring
Example

• Prospect Scoring
• 5 Million Customer Base
• Initial Sample 50,000
• 1000 Responded (2)
• Need to Increase Response to 4
• Information Gathered
• Time Acct. Open (TIMEAC)
• Avg. Acct. Balance (AVEBAL)

16
Prospect Scoring Example
Respondents
Time Account Open, Years
Average Account Balance
17
Prospect Scoring Example
Non-Respondents
Time Account Open, Years
Average Account Balance
18
Neural Networks in MarketingProspect Scoring
Example

• 2 Inputs (TIMEAC AVEBAL)
• One Output (Score)
• Zero for respondent
• One for non-respondent
• Training Set (1000 Cases)
• 500 Randomly from Resp. Pool
• 500 Randomly from Non-resp. Pool

19
Prospect Scoring Example
NN Model Fit
20
Neural Networks in MarketingProspect Scoring
Example

• Test Set (1000 Cases)
• Reaming 500 respondents
• 500 Randomly from Non-respondents
• Predict as Gains Chart
• Model Calculates SCORE for Test Set
• Rank in Descending Order of SCORE
• Add-up No. of Resp. Non-resp.
• Plot on the Chart

21
Prospect Scoring Example
NN Model Gains Chart
Cumulative Response, ()
Cumulative Non-response ()
22
Neural Networks in MarketingProspect Scoring
Example

• Results
• Mail Top 1 Million Savers
• Generate 40,000 New Prospects
• Cost Saving - 300000 Pounds
• Conclusion
• Alternate to Statistical Tools
• Handle Large Amounts of Data

23
NN Applications in Finance
NN for Managing Investments

• Issues to Consider
• Optimal Time Horizon
• SPIX into 10 days Forward
• Input Set to Use
• Change in the Indicator
• Periodicity of Indicator
• How Much Lag

24
NN for Managing InvestmentsExample
Prediction of SP 500 Index Futures

• Inputs - 21 indicators
• CRB Index, Index, Etc. (7)
• Price, Volume, Put-Call Ratio (21)
• Time Lags
• 5, 10, 15, 20, 25
• Total of 126 Variables
• Predict 5 days forward

25
SUMMARY

• ANNs Universal Function Approximators
• Even for non-linear functions
• Can handle discontinuities
• Estimate Piece-wise Approximations
• Trigger Use Specialized Models
• ANNs can be automated
• ANNs learn incrementally

26
SUMMARY (contd..)

• Changing Technology
• ANN methodology changing
• Interpretation
• Hard to interpret/ Physical meaning
• Number of Parameters
• ANNs usially have more /Overfitting
• ANNs need more computer power