You are on page 1of 7

Ensemble model for classifiers

Submitted by : Group 4 Gandhrav Himanshu Navneet Shanay 1


Data Set
Data of payment over past 6 months of a customer is given

There are 23 variables used to predict weather a customer will default on its payment or not

Attribute information
X1 X2 X3 X4 X5 X6 – X11 X12 – X17 X18 – X23
Amount of the Gender Education Marital Age History of past Amount of bill Amount of previous
given credit status payment statement payment

*Source: UCI Machine Learning Repository (Link) 2


Classifiers
Classifiers

Logistic K – Nearest
Decision Tree
Regression Neighbor (KNN)

 Decision tree builds classification It is a statistical method for k-Nearest Neighbor is a lazy learning
or regression models in the form analyzing a data set in which there algorithm which stores all instances
of a tree structure are one or more independent correspond to training data points in
 The tree is constructed in a top- variables that determine an n-dimensional space
down recursive divide-and- outcome
conquer manner

3
Classification Method Results
Decision Tree Logistic Regression KNN Clustering
Predicted Predicted Predicted
0 1 0 1 0 1
Actual 0 34 3 Actual 0 34 3 Actual 0 32 5
1 7 6 1 8 5 1 10 3

Accuracy = 80% Accuracy = 78% Accuracy = 70%

4
Ensemble Model
• Majority Voting
• We assigned the prediction for the observation as predicted by the majority of models

• Weighted Average
• Instead of taking simple average, we took weights based on the accuracy of the three models

Model Decision Tree Logistic Regression KNN clustering


Accuracy 80% 78% 70%
Weights 0.6 0.2 0.2

Ensemble Model Accuracy


Majority Voting 78%
Weighted Average 80%

5
Applications
•Fraud Detection • Identification of bank frauds such as money laundering or credit card default

• Can predict financial crises and financial distress


Financial Decision Making • Help to analyze changes in stock market data which can stop trade based
manipulation

Face Recognition • Identification or verification of a person using his/her digital images

• Speech based emotion recognition is done using the ensemble


Emotion Recognition learning model

Computer Security / Data • Identification of malware or unusual activities


Privacy

6
Thank you

You might also like