SONAR Object Classifier

This project implements a basic classification algorithm using a perceptron model created from scratch, in order to classify objects found in a SONAR dataset, into Mines & Rocks.

Pseudocode of Perceptron Algorithm

1. Initialize the weights and threshold to small random numbers.
2. Present a vector X to the neuron inputs and calculate the output
3. Update the weights according to:

   weights[i+1] = weights[i+1] + learning_rate * error * row[i]

   where
   • i is the iteration number, and
   • learning_rate is the gain or step size, where 0.0 < n < 1.0
4. Repeat 2 & 3 until:
   • The iteration is less than a user-specified error threshold or
   • The predetermined number of iterations have been completed

Output

('M' denotes Mine, 'R' denotes Rock)
Output of the Perceptron Algorithm for the SONAR dataset with the following values

• Folds : 2
• Initial Learning Rate : 0.01
• Epochs : 400

The output here is taken with the same values for folds, learning_rate and number_of_epochs but the accuracy varies beacuse the Cross-Validation technique makes folds at random.

{'M':0, 'R':1}

Scores : [83.65384615384616, 69.23076923076923]
Mean Accuracy : 76.442 %

Scores : [72.11538461538461, 68.26923076923077]
Mean Accuracy : 70.192 %

Scores : [70.1923076923077, 74.03846153846155]
Mean Accuracy : 72.115 %

Acknowledgments

Thanks to the code provided by Matthew Carter. The code used as reference for the project is available here.