quintenmm / us-income

using ml to predict income of us household. Than optimizing the model to prevent over fitting

US-Income challenge

In this assignment we are given 2 datasets. one for training and on for testing. The goal is to predict income of us-citizens based 14 features.

First we measure baseline accuracy using random forrest.

after this we'll try to improve the score by tuning hyperparameters and try to compensate for overfitting.

baseline accuracy

Classifier model	Accuracy score	Set type	Random state
RandomForrestClassifier	99,99%	train	30
RandomForestClassifier	85,14%	test	30

as you can see traing score is very high but drops when given unseen data. this is a clear sign of overfitting. In the confusion matrix below we see a lot of data is misclassified as false negatieve.

cross-validation

cross validation was my first attempt at getting a better score. I fitted the traing set and got a lower accuracy score. This combets overfitting

cross validation score	number of folds	standard deviation
85.72%	10	+/- 0.45

Mission objectives

• Be able to analyze a machine learning problem
• Be able to reason about possible causes of overfitting
• Be able to remedy the causes of overfitting
• Be able to tune parameters of a machine learning model
• Be able to write clean and documented code.

conclusion

I kept seperating features and refitting to get a better score. The dataset is inbalenced from the beginning. Even if i rebalence and keep looking for better features i can't get the score higher than an extra 2%. using more trees and optimising the model gives deminishing returns after severel folds.

Contributors

Name	Github
Quinten Wildemeersch	https://github.com/QuintenMM