LEARNING OBJECTIVES
Define neural networks.
Identify and compare various uses for neural networks.
Explain the elements of a neural network.
Investigate how neural networks function.
Review important considerations when using neural
networks
Introduction to Neural Networks
Which customers are likely to purchase a product or
service following exposure to a digital marketing
campaign?
How much will a customer spend? These types of
prediction or forecasting questions can be examined
using a number of methods for analyzing data.
For example, answers to some of these questions might
be obtained by using predictive linear regression.
Introduction to Neural Networks
Regression is the appropriate method when the
predictive relationships are linear.
Linear regression could identify the relationship
between the number of times a customer visits a
website and whether they are likely to make a
purchase.
Introduction to Neural Networks
But other relationships might be nonlinear in
nature, such as examining seasonal visits to a
website and the likelihood of a purchase.
In this situation, a non-linear method such as
neural networks can provide more accurate
answers to these questions. Neural networks are
algorithms that are trained to recognize patterns in data
that are non-linear.
The logic of neural networks is based on patterns
identified by observing biological activities in the
human brain
Introduction to Neural Networks
Brains have networks of neurons (connections)
that transfer information. The neurons learn from past
experiences and use this knowledge to solve problems.
For example, a brain recognizes a chair is in fact
something to sit on based upon the learned
knowledge of selected experiences.
A similar process occurs within a neural network. That
is, relationships (networks) between data, and
the learned patterns are applied to solve complex
problems.
How Are Neural Networks Used in
Practice?
Marketers use neural network technology to identify
patterns and make accurate predictions.
Neural networks can aid marketers in predicting
customer behavior, understanding buyer
segmentation, developing brand strategies,
optimizing inventory, forecasting sales, improving
marketing automation, developing digital content,
and much more.
How Are Neural Networks Used in
Practice?
Classify new customers by their potential profitability when
planning direct marketing strategies
determine which customers are most likely to open their
direct mail based on past purchase behavior
Generate customized digital content
Predict lead scoring.
Personalize customer experiences.
Determine customer lifetime value and defined customer
segments.
Increase new customers that request online product quotes
Basic Terminology of Neural
Networks
Basic Terminology of Neural
Networks
The overall neural network represents a classification
problem.
The purpose is to classify the customers into two groups:
purchasers and non-purchasers.
Neurons or nodes: A set of inputs (features) that are
multiplied by their weights. To do so, the values are
summed and applied to what is referred to as an
activation function. The result is to convert the inputs into
a single output.
Basic Terminology of Neural
Networks
Inputs: Values representing features (variables) from the
dataset that pass information to the next layer via connections.
Connections: Transmit information from one neuron to
another. Each network consists of connected neurons and each
connection is weighted.
Weights: Determines how important each neuron is.
Connections between the neuron layers are assigned a weight.
Features that are important in predicting the output have
larger weights. In contrast, less important features have
smaller weights. Weights are produced in the hidden layers.
Basic Terminology of Neural
Networks
Learning weight: Determines the amount of
adjustment made to the weights in the network.
Learning rate: Determines the speed at which the
model can arrive at the most
accurate solution. The learning rate consists of
positive values that generally range between 0 and 1.
Hidden layer: Calculations are performed in the
hidden layer to produce weights from the input layer.
The weights are then transferred to the next hidden
layer or output layer.
Basic Terminology of Neural
Networks
Outputs: Where the model arrives at a prediction. The
output is calculated as:
Forward propagation: Process whereby data from the
input layer moves forward through the network. The
data is sent through the hidden layer where it is
processed using an activation function. The data
produced is again sent forward to the next layer.
Basic Terminology of Neural
Networks
• Backpropagation: Transmits the total loss back into the
neural network to understand the amount of loss from each
neuron. Using this information, weights are then adjusted to
reflect the loss. Nodes with higher error rates are given
lower weights, and those with lower error rates are given
higher weights.
• Biases: The constant value given to the weighted input of
each node. This constant value changes the weighted sum
of the inputs, which in turn, adjusts the output. If bias is
absent, then the neuron might not be activated. As a result,
the neuron’s information would not be passed to the next
layer.