Half-Moon Dataset Classification via MATLAB
Budget: $30 – $250 USD
MATLAB - Back Propagation Neural Network
apply a back propagation neural network to implement pattern classification for the linearly nonseparable half-moon dataset (the first 1,000 samples belong to Class 1, and the second 1,000 samples belong to Class 2). The network consists of two input neurons, one hidden layer with an adjustable number of hidden neurons, and one output neuron.
1. Online learning algorithm
- Plot the learning curve (number of epochs versus mean squared error for each epoch).
- Evaluate the classification accuracy for the training set and the entire dataset.
- Investigate the impact of the learning rate.
- Investigate the impact of the steepness parameter of the activation function.
2. Batch learning algorithm
- Plot the learning curve.
- Evaluate the classification accuracy of the training set and the entire dataset.
- Investigate the impact of the number of hidden neurons.
- Investigate the impact of data preprocessing via normalization (scaling each input
vector to unit length).
Matlab code must be fully and extensively commented.
apply a back propagation neural network to implement pattern classification for the linearly nonseparable half-moon dataset (the first 1,000 samples belong to Class 1, and the second 1,000 samples belong to Class 2). The network consists of two input neurons, one hidden layer with an adjustable number of hidden neurons, and one output neuron.
1. Online learning algorithm
- Plot the learning curve (number of epochs versus mean squared error for each epoch).
- Evaluate the classification accuracy for the training set and the entire dataset.
- Investigate the impact of the learning rate.
- Investigate the impact of the steepness parameter of the activation function.
2. Batch learning algorithm
- Plot the learning curve.
- Evaluate the classification accuracy of the training set and the entire dataset.
- Investigate the impact of the number of hidden neurons.
- Investigate the impact of data preprocessing via normalization (scaling each input
vector to unit length).
Matlab code must be fully and extensively commented.