Classification Using Artificial Neural Networks

Introduction

During a data mining class for the university, an investigation and development (I+D) activity encourages me to work as a team with two other students to conduct research on neural networks in general, develop an approach and present our results to the rest of the class.

Our implementation part was the development of a 60 input unit, 60 hidden units and 1 output unit artificial neural network for a classification task using this UCI dataset.

Conclusions

The results of this project are the comprehension of essential aspects of neural networks such as layers, perceptrons, backpropagation, activation and loss functions, optimizers, etc.

Also the implementation of those concepts using Python + Keras, building an ANN with 80% accuracy.

Skills used in the final solution:

• Sequential fully connected neural network structure.
• Sigmoid as the ANN activation function .
• Adaptative momentum (Adam) optimizer.
• Binary cross-entropy as loss function to train artificial neural networks for classification .