A neural network classifier for the Reuters news dataset using Keras.
This repository contains a neural network implementation for classifying Reuters news articles into 46 different categories. It uses a simple yet effective architecture with two dense hidden layers to achieve multi-class classification of news content.
The classifier demonstrates:
- Text data preprocessing and vectorization
- Multi-class classification with neural networks
- Training visualization and evaluation
- Automatic loading and preprocessing of the Reuters dataset
- Text vectorization using one-hot encoding
- Neural network model with configurable architecture
- Training and validation metrics visualization
- Multi-class classification capability
- Python 3.6+
- TensorFlow 2.x
- Keras
- NumPy
- Matplotlib
-
Clone this repository:
git clone https://github.com/corticalstack/ANNReutersClassifier.git cd ANNReutersClassifier -
Install the required dependencies:
pip install tensorflow numpy matplotlib
-
Run the classifier:
python main.py
The neural network architecture consists of:
- Input layer accepting 10,000-dimensional vectors (bag-of-words representation)
- First hidden layer with 64 neurons and ReLU activation
- Second hidden layer with 64 neurons and ReLU activation
- Output layer with 46 neurons (one per category) and softmax activation
The model is compiled with:
- RMSprop optimizer
- Categorical crossentropy loss function
- Accuracy metric
- Data Loading: The Reuters dataset is loaded with a vocabulary limited to the 10,000 most frequent words.
- Text Preprocessing: Each news article is converted into a fixed-length vector using one-hot encoding.
- Label Encoding: The category labels are one-hot encoded.
- Model Training: The neural network is trained for 20 epochs with a validation split.
- Visualization: Training and validation loss/accuracy are plotted to evaluate model performance.
This project is licensed under the MIT License - see the LICENSE file for details.