Brief Introduction

The project is an Image Classification Model for Symbol Reorganization.

The project goal is to achieve at least 90% of classification accuracy. Weka is the machine learning software used for this project.

Steps

1. Create datasets, which includes a training set and testing set.

2. Feed the training data into Weka, train the model with default parameter settings.

3. Fine tuning the model.

Outcomes

With default parameter settings, the neural network has 3 layers, and the model accuracy is 88.8462%, which is lower than my goal.

After fine tuning the model by experimenting with hidden layers, activation function, loss function, learning rate, time of iterations, and number of neurons.

My final optimized model has training data accuracy of 98.0796%, and testing data accuracy of 93.3333%.

It is a 4 layer neural network, 30 times iteration. In the added hidden layer, there are 4 neurons, using ‘Cube’ as the activation function . In the output layer, using ‘Softmax’ as the activation function, and ‘LossConiseProximity’ as the loss function.