Scribd
Upload
102 views5 pages

ML Interpretability Assignment

This document discusses machine learning interpretability and provides recommendations for improving the interpretability of models. It ranks models from most interpretable to least interpretable as linear regression, logistic regression, decision trees, random forests, and neural networks/deep neural networks. It suggests using feature importance to identify key predictive variables and reduce complexity. The LIME method and Shapley values are also proposed to help explain predictions from "black box" models and determine the contribution of features.

Uploaded by

Krishna Priya Gazula
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
102 views5 pages

ML Interpretability Assignment

This document discusses machine learning interpretability and provides recommendations for improving the interpretability of models. It ranks models from most interpretable to least interpretable as linear regression, logistic regression, decision trees, random forests, and neural networks/deep neural networks. It suggests using feature importance to identify key predictive variables and reduce complexity. The LIME method and Shapley values are also proposed to help explain predictions from "black box" models and determine the contribution of features.

Uploaded by

Krishna Priya Gazula
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 5

Machine Learning Interpretability Assignment

Machine Learning Concepts

Activity : Group Assignment


Topic : ML Interpretability

Submitted by,
F20022 – G.L. Krishna Priya
F20077 – Ashwin.V
F20102 – Melvin Jose P G
F20117 – Souradeep Pal
F20139 – J R Dinesh
Machine Learning Interpretability Assignment

1. Which are the models that can explain the features better? Which models cannot
explain better?

The models that can explain features better and are more interpretable are:

 Linear Regression

 Logistic Regression

 Decision Trees

 Support Vector Machines.

When the reasoning behind the predictions and decisions readily understandable, it is
considered interpretable.

If the model is linear, monotonic and the output is available at the defined rate it also
helps. The 3 methods are also more interpretable because there is a strong mathematical
base amongst the models. All are steeped in mathematics and mathematical equations
in calculations aside from being readily understandable.

The models that cannot explain features better are:

 XG Boosting

 Gradient Boosting Method

 Random Forests

 Neural Networks

 By extension, Deep Neural Networks.

For random forests, the no. of trees can range up to a large amount and this can make the
process far more complicated as opposed to Decision trees. Likewise, the same can be
said be said for Neural Networks and Deep Neural Networks where the no. of neurons can
range to a high amount(no. of weights). XGBoosting and Gradient boosting are similar.
Machine Learning Interpretability Assignment

There are no fixed values or mathematical calculation attached to these methods(no. of


trees and no. of neurons vary and the calculations can very accordingly).

2. Rank the Models based on which can explain the features better.

The following Rank would be assigned to the list of models.

1. Linear Regression

2. Logistic Regression

3. Decision Tree

These models explain features well. Linear Regression is not only linear and monotone but
the output changes at the defined rate.

The mathematical calculations are done based on the y=mx + C formula to determine how
much result(Y) will depend on a variable. Logistic Regression uses the logit function and
is also based on calculations and mathematics which increases interpretably. Decision Tree
is easy to understand and follows Sum of Product.

Also cost of each option along the decision line is calculated for Decision tree analysis.

4. Random Forest

5. Neural Network and Deep Neural Network.

Random Forest is ranked low because of the no. of trees within the forest not being fixed
and the calculations not being fixed and steeped in math as well.

Neural Network is more advanced with the no. of weights being very high in addition to
calculation complexities caused by the neurons.
Machine Learning Interpretability Assignment

3. How can we address the issue of ML interpretability(especially for those models


which rank the least because of their expression capabilities)?

a) Feature Importance:

We can use Feature of Importance. We can thus use this to see the contribution of
variables and see which variables explain the accuracy better.

Often, it is found that around 20% of the variables explain the accuracy very well as
opposed to the rest 80%. We can use this to just select the important variables and
reduce the unnecessary complexity and increase interpretability by making it more
understandable.

b) LIME Method: We can also use the LIME Method (Locally Interpretable Model
agnostic Explanation). This is a method used for black-box models (such as the lower
ranked ones). Here we look at the subsets of data instead of the whole data and
evaluate. LIME explains a prediction so that non-experts can compare and improve on
a model.

It explains the prediction of any classifier by learning an interpretable model locally


around the prediction. It creates a representative sample set and LIME provides a
global view of the model's decision boundary to its users. It is crucial for human
interactions with Machine Learning systems.

c) Sharpley Value: We can use the Sharpley value. It is the average expected marginal
contribution of a single player after every possible combination is taken into account.
Machine Learning Interpretability Assignment

It determines the payoff for all the players when each might have contributed more or
less than the others. It is used in Game Theory. It finds out the contribution of the
players.

So first, the Model is built, then after each iteration the feature importance is assigned.
The features will be explained better and bias scripts will be taken care of in the
model(removal of systemically prejudiced results due to wrongful assumptions).

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy