A Review on Machine Learning Techniques for Fake News Detection

Abstract—Spreading false news online has become a grave
threat to public stability, political life, and trust. Conventional
content moderation measures prove inadequate to contain the
velocity and volume with which disinformation spreads. Machine
Learning (ML) has been an invaluable aid in the automation of
fake news identification by learning patterns and cues from large
amounts of text data. This review paper provides an overview
of most of the ML methods employed for the detection of
fake news, from the simple algorithms Naive Bayes, Support
Vector Machines, and Decision Trees to the recent deep learning
methods such as LSTM, CNN, and transformer-based models
such as BERT. This work also encompasses feature extraction
methods like TF-IDF, word embeddings, and stylometric features,
benchmarking datasets, and evaluation metrics used in the
literature. The work provides the advantages and limitations
of existing methods and highlights open problems such as
dataset generalizability, model explainability, and adversarial
robustness Lastly, the review outlines future directions, such as
the development of hybrid models, real-time detection systems,
and ethics. The intention of this paper is to be a starting ground
for researchers as well as practitioners in creating improved and
more efficient fake news recognition systems.
Index Terms—Fake News Detection, Machine Learning, Natural
Language Processing, Text Classification, Deep Learning,
Transformer Models, Social Media Misinformation, NLP, News
Verification