Determine which class - fake or credible - some observation probably belongs to by using the Bayes formula.
Dataset - fake news
Steps taken to perform the task :
- Analyze the dataset, process it to turn into a bag-of-words
- Use Bayes' formula to try to predict the class of a message
- Calculate the probability for the bag-of-word to be in fake class
- Calculate the probability for the bag-of-word to be in credible class
- Compare them
- Compute the success of the predictions
- Calculate the metrics to evaluate the effectiveness of the classifier
- Form the confusion matrix to represent the effectiveness of the method
- Form the diagram and visualize the statistics
Accuracy : approximately 93% (0.9289 out of 1)
Pros of using Naive Bayes approach :
- Simple and easy to implement. Naive Bayes uses basic probability formulas and concepts and clear algorithm.
- Comparatively fast by the assumption that the features are independent. That is one of the reasons why Naive Distribution is used on big datasets.
Cons :
- Does not work for non-independent features. The count of datasets having completely independent features is reaching zero, so we can use Naive Bayes only when speed is more praised than the accuracy
In the dataset given, we actually assumed that there was no correlation between the word frequency and our method still had good results. However, it could be a good practise e.g to consider the dependencies of the word appearance. Still, Naive Bayes has pretty good classification accuracy.
The MIT License (MIT)
Copyright © 2022. Andrii Yaroshevych, Dmytro Vasylkiv, Vitalii Petrychko