Sei sulla pagina 1di 18

BUSINESS INTELLIGENCE AND

ARTIFICIAL NEURAL NETWORK (ANN)


THANIA
BUSINESS INTELLIGENCE
BUSINESS INTELLIGENCE SYSTEM

 Organizing and analyzing data into information that will help a company to make better decisions faster.

 BI cycles:
 Gathering data
 Turn data into information
 Make decisions based on information from data
 Review the results related to the company’s key performance indicator (KPI)
 Make the next decisions from updated data and reviews from previous decision
BUSINESS INTELLIGENCE SYSTEM

 To make better decisions, a company needs:


 Data - nominal, ordinal, interval, ratio, BLOBs (Binary Large Obejct: images, audio, video, etc.)
 Database - operational / transactional database
 Data warehouse - long term storage for data
 Data mining - decision trees, regression, ANN, cluster analysis, association rule mining
 Data visualization - graphs, charts, dashboards
DATA MINING
ARTIFICIAL NEURAL NETWORK (ANN)
ARTIFICIAL NEURAL NETWORK (ANN)

 ANN is one of data mining techniques

 ANN can be used for many kinds of pattern


recognition and prediction.

 ANN is developed by mimicking neurons in the


human brain or nervous system

 ANN is trained through a learning process (training


and testing)
ARTIFICIAL NEURAL NETWORK (ANN)

 Constant = 1 and W0 (bias weight) are there


to move the output function of each
perceptron in the number graph
7 COMMON ACTIVATION FUNCTIONS

 Sigmoid function
 TanH function
 ReLu function
 Leaky ReLu
 Parametric ReLu
 Softmax
 Swish function

 Experimenting with different activation functions can lead to better results


BIAS AND VARIANCE IN NEURAL NETWORK

 Training set - bias


 Testing/validation set - variance

 Error function: checking how far away the prediction from the
true value
BACKPROPAGATION

 Backpropagation algorithm calculates how much the output values are affected by each of the weights in the
model.
 Calculates partial derivatives
 Weight update
 After every sample in training set < after every batch of training set

 Hyperparameter
HYPERPARAMETERS IN NEURAL NETWORK

 Number of hidden layers


 Activation function
 Weight initialization
 Learning rate
 Batch size
 Etc.
BASIC ANN ALGORITHMS

 Feed-forward algorithm

 Activation function (sigmoid)

 Error function
 Backpropagation

 Updated weight
ADVANTAGES OF ANN

 ANN has very little restrictions


 ANN can be used even when there are no algorithmic solutions or when algorithmic solutions are too complicated.
 ANN does not need to be programmed
 ANN gets better the more it learns / used.
 ANN can handle many types of problems
 ANN can be used for many kinds of pattern recognition and prediction
 ANN can handle numerical and categorical variables
 ANN does not have data restrictions
 ANN can produce better result if it is trained enough
DISADVANTAGES OF ANN

 ANN needs expertise and a lot of experimentation to reach optimal design


 It can be difficult to handles large number of variables with ANN
 ANN needs large sets of data to train
 Training set (to adjust input weights)
 Testing/validation set (to verify the accuracy)
DIFFERENT TYPES OF ANN
REFERENCES

 Gershenson, Carlos. n.d. Artificial Neural Networks for Beginners


 Maheshwari, A. K. (2015). Business Intelligence and Data Mining. New York: Business Expert Press
 https://missinglink.ai/guides/neural-network-concepts/complete-guide-artificial-neural-networks/
 https://towardsdatascience.com/step-by-step-guide-to-building-your-own-neural-network-from-scratch-
df64b1c5ab6e
 https://opendatascience.com/essential-neural-network-algorithms/
 https://www.digitalvidya.com/blog/types-of-neural-networks/

Potrebbero piacerti anche