"Multilayer perceptron (MLP) is a technique of feed forward artificial neural network using back propagation learning method to classify the target variable used for supervised learning. It consists of multiple layers and non-linear activation allowing it to distinguish data that is not linearly separable."

1. Master the Art of Analytics
A Simplistic Explainer Series For Citizen Data Scientists
J o u r n e y To w a r d s A u g m e n t e d A n a l y t i c s

2. Multilayer Perceptron Classifier

3. Terminologies
Introduction & Example
Standard input/tuning parameters & Sample UI
Sample output UI
Interpretation of Output
Limitations
Business use cases
What is
covered

4. Terminologies
▪ Target variable usually denoted by Y, is the variable being predicted and is also called dependent variable,
output variable, response variable or outcome variable (E.g., One highlighted in red box in table below).
▪ Predictor, sometimes called an independent variable, is a variable that is being used to predict the
target variable (E.g., Variables highlighted in green box in table below).
The predictors highlighted in green box above constitutes of the attributes upon which the target variable
highlighted in red box (i.e., Opportunity Result) depends on.
Opportunity
result
Revenue from
client past 2
years
Total days
identified through
qualified
Total days
identified
through closing
Ratio days
qualified to
total days
Sales stage
change count
Won 3 52 117 0.30316 17
Loss 0 74 74 0.896505 9
Loss 0 115 115 0.0 3
Loss 0 80 80 0.0 3
Won 0 29 29 0.0 7

5. Terminologies (Continued…)
• Layers:
• Multilayer Perceptron consists of three layers, input layer, hidden layers and output layer. Layers take
array as the input and each value in the array represents the size for the three layers.
• Weights:
• Weights control the strength between the two nodes, i.e., they specify how much influence will the
input have on the output.
• Feed Forward Neural Network:
• Multilayer Perceptron model is a feed forward network because information flows through the
function being evaluated from the input layer, through the intermediate computations, and finally to
the output, i.e., all the layers are fully connected and the information flows from one layer to the
other.
• Backward Propagation:
• A procedure to repeatedly adjust the weights so as to minimize the difference between actual output
and desired output.

6. Introduction
• Objective:
͢ Multilayer perceptron (MLP) is a technique of feed
forward artificial neural network using back propagation
learning method to classify the target variable used for
supervised learning.
• Benefit:
͢ MLP’s can be applied to complex non-linear problems,
and it also works well with large input data with a
relatively faster performance. The algorithm tends to
achieve the same accuracy ratio even with smaller data.
• Model:
͢ In the Multilayer perceptron, there can be more than one
linear layer. For instance, a 3-layer network will have the
first layer as the input layer, middle layer as the hidden
layer and last layer as the output layer. We can feed data
into the input layer and get the classification output from
the output layer and increase as many hidden layers as
required to cater to the complexity of the task.

7. Example: Multiple Perceptron Classifier
Independent
variables (Xi)
Target
Variable (Y)
Let’s conduct the Multilayer Perceptron Classifier analysis on independent variables: Revenue, Total days(qualified), Total
days(closing), Ratio days, Sales Stage and target variable: Opportunity Result as shown below:
Opportunity
Result
Revenue from
client past 2
years
Total days
identified
through
qualified
Total days
identified
through
closing
Ratio days
qualified to
total days
Sales Stage
Change
Count
Won 3 52 117 0.303 17
Loss 0 74 74 0.896 9
Loss 0 115 115 0.0 3
Loss 0 80 80 0.0 3
Won 0 29 29 0.0 7
Model is an
excellent fit as
Accuracy > 75%
• Classification Accuracy:
○ A crucial criterion for assessing Model
Performance
○ Model with prediction accuracy > 75% is
useful.
• Classification Error = 100- Accuracy = 14.52%
○ Indicates that there is 14.52% chance of
error in classification
Classification Evaluation Metric
Accuracy 85.48%
Classification Error 14.52%

8. Standard Input/Tuning Parameters & Sample UI
Step
1
Step
2
More than one
predictors can be
selected
Step 3
Block Size = 128
Maximum number of
Iterations = 100
By default, these parameters
should be set with the values
mentioned
Step 4
Display the output window containing following:
● Scatter Plot
● Dimension Contribution
● Dimension Counts By Percentage
● Average Measures by Target Classes
Note:
▪ Decision on selection of predictors depends on the business knowledge and the correlation value between target variable and predictors.
Select the Target Variable
Opportunity Result
Revenue from client past 2 years
Total days identified through qualified
Total days identified through closing
Ratio days qualified to total days
Sales Stage Change Count
Select the Target Variable
Opportunity Result
Revenue from client past 2 years
Total days identified through qualified
Total days identified through closing
Ratio days qualified to total days
Sales Stage Change Count

9. Sample Output: 1. Interpretation
Influencer’s importance chart is used to show impact of each predictor on target variable.
Target Variable: Opportunity Result
Influencer’s Importance

10. ● Accuracy: It shows the goodness of fit of the model. It lies
between 1 to 100 and closer the value to 100, better the model.
● Precision: Proportion of predicted values that were actually correct. Generally, higher precision (>70%) indicates
that confidence for predicted class is high.
● Recall/Sensitivity/Hit Rate: Proportion of actual positives that were predicted correctly. Generally, higher recall
(>70%) indicates that confidence for predicted class is high.
Precision Recall
Loss 89.66% 91.88%
Won 69.27% 63.31%
Accuracy 85.48%
Class Wise Precision and Recall
Predicted
Loss Won
Actual
Loss 3476 307
Won 401 692
Actual versus Predicted Class
Sample Output: 2. Model Summary

11. Sample Output: 3. Predicted Class & Probability
Opportunity
Result
Revenue
from client
past 2
years
Total days
identified
through
qualified
Total days
identified
through
closing
Ratio days
qualified to total
days
Sales Stage
Change Count
Probability Predicted
Opportunity Result
Won 3 52 117 0.303 17 0.78 Won
Loss 0 74 74 0.896 9 0.92 Loss
Loss 0 115 115 0.0 3 1.0 Loss
Loss 0 80 80 0.0 3 1.0 Loss
Loss 0 29 29 0.0 7 0.92 Loss
The data output will contain predicted class column along with the probability of prediction

12. RMSE R Squared
RMSE R-Squared
Accuracy: Precision: Recall:
• Accuracy > 75% represents
model is well fit on the
provided data and the
values are reasonably
accurate.
• Accuracy < 75% represents
model is not well fit on
provided data and the
values are likely to be
inaccurate and contain high
chances of error.
• Proportion of predicted values
that were actually correct.
Generally, higher precision
(>70%) indicates that
confidence for predicted class is
high.
• Proportion of actual
positives that were
predicted correctly.
Generally, higher recall
(>70%) indicates that
confidence for predicted
class is high.
Interpretation of Important Model Summary Statistics

13. Interpretation of Plots: Scatter Plot
● This plot is used to see the classification quality by model; the less overlap among the classes in
the plot above, the better the classification by model.
● We can also visually analyze how a particular class is assigned.
● Scatter plots give the overview of the input data, allowing a user to see general trends for the
attributes.
● The graph is plotted against measures within the data.
Sales
Stage
Change
Count
Ratio Days Qualified to Total Days
Won Loss

14. Interpretation of Plots: Dimension Contribution
● This plot is used to display how dimension values are distributed for each class in the target variable.
● For instance, the plot above shows how various values Supply Groups such as (Car Accessories, Car
Electronics, Performance & Non - Auto, Tires & Wheels) that are distributed within each target class
(Won, Loss). The graph shows counts of target class(Won, Loss) for the predictor classes chosen.

15. Interpretation of Plots: Dimension Counts by Percentage
● This plot is used to visually analyze how dimension counts are distributed across target variable classes.
● For instance, the plot shows how various supplies group are distributed within each opportunity result
class to analyze whether or not a particular target class has is having relatively more counts of particular
supplies group segment

16. Interpretation of Plots: Average Measures by Target Class
• This plot is used to visually analyze how average measures are distributed across target variable classes.
• For instance, the plot above shows how different predictor measure variables are distributed within each
Opportunity Result (Won, Loss) class.
Avg(Revenue from client
past 2 years)
Avg(Total days identified
through qualified)
Avg(Total days identified
through closing)
Avg(Ratio days qualified to
total days)
Avg(Sales Stage Change
Count)
Average
Opportunity Result

17. Limitations
● The extent to which an independent variable is affected by dependent variable is unknown
and thus its computations are difficult and time consuming.
● Minimum of 1000 data points are required to get reliable predictions.
● The quality of training must be good in order to ensure the proper functioning of the
model.
● Multilayer Perceptrons include too many parameters since it is fully connected, i.e., each
perceptron is connected to every other leading to growth in total number of parameters
causing information redundancy in higher dimensions.

18. Limitations (Continued…)
● A normal distribution is an arrangement of
a data set in which most values cluster in
the middle of the range and the rest taper
off symmetrically towards extreme. It will
look like a bell curve as shown in figure 1.
● Outliers in data (target as well as
independent variables) can affect the
analysis, hence outliers need to be
removed.
● Outliers are the observations lying outside
overall pattern of distribution as shown in
figure 2.
Figure 1
Figure 2

19. Business Use Case 1
• Business problem: Predict employee attrition
• Identifying the important factors that lead to the employee attrition.
• Input data:
• Predictor/independent variables:
• Overtime
• Monthly Income
• Total Working Years
• Stock Option Level
• Relationship Satisfaction
• Target/dependent variable:
• Attrition
• Business benefit:
• The predictive model will help us identify various factors that affect the resignation or retirement
decisions made by the employee. This will help the companies identify the criteria's it needs to work
on to retain employees in the company.

20. Business Use Case 2
• Business problem: Predicting medication type for patients in a hospital
• Identifying the right type of medication/treatment for various patients admitted in the hospital
• Input data:
• Predictor/independent variables:
• Time Spent in Hospital
• Number of Medications
• Number of Procedures
• Patient’s Weight
• Medical Specialty Ward
• Target/dependent variable:
• Target (Drug, Solo Insulin)
• Business benefit:
• Filtering through the most important factors of a patient’s diagnosis to help choose the most
appropriate type of medication (Drug, Solo Insulin) for the patient.

21. Want to
Learn More?
Get in touch with us @
support@Smarten.com
And Do Checkout the Learning section
on
Smarten.com
September 2021