This is a Natural language processing project which identifies if a news is FAKE or REAL.
The datset used contains 6335 unique news articles with FAKE/REAL labels for each with no null values.
- LSTM architecture.
- Tokenizer
The loss value of 0.0290 has been achieved which gives good results but the model is not generalised and will give a lower accuracy on data input from outside the testing data.
Training and testing data was tokenised with appling num_words limited to 2000 and max_len fixed at 400, and created a padded sequence.
Model evaluation resulted in 0.8245 accuracy metric.
- Pandas
- Matplotlib
- Sklearn
- Keras