Team Batch Size

Day 2: Classification

Classification is a type of supervised learning that determines what class an object belongs to.

In our data, we classified images by emotion; each image is an object, and each emotions is a classes.

Decision Tree

K-nearest neighbours

Logistic Regression

Random Forest

Makes decisions by splitting up the training set based on specific features

Called a rule-based classifier because each step follows a specific rule to make a decision

Uses supervised learning and is very sensitive to training data

With 34% accuracy, this was the least accurate classification method for our data.

training (left) and validation (right) results while using decision trees

Uses supervised learning.

Classifies a data point by finding similar characteristics with other labeled points.

Requires parameter tuning, which is the process of choosing the right value of k; we used a K value of 3.

At 36% accuracy, this is the second least accurate classification method for our data.

training (left) and validation (right) results while using the k-nearest neighbour classification.

Logistic regression uses a logistic sigmoid function to fit the data.

Y is either 0 or 1, which classifies the data into binary sets.

With 47% accuracy, this is one of the most accurate classification method for our data.

training (left) and validation (right) results while using the logistic regression classification.

Consists of many decision trees

Uses bagging and feature randomness when building each tree

The best prediction is chosen after each tree has provided a conclusion

With 47% accuracy, this was one of the most accurate classification method for our data.

training (left) and validation (right) results while using random forest classification.

Bagging

Using different sets of training data for different trees

Feature Randomness

Using different features to make decisions for different trees

Page updated

Google Sites

Report abuse