Perceptron Network for Iris Dataset Classification

A Perceptron-based neural network implementation for multiclass classification of the Iris dataset using a one-vs-all strategy

This repository contains an implementation of a Perceptron Network designed to classify the Iris dataset into its three species (Iris-setosa, Iris-versicolor and Iris-virginica). The system is based on the classical Rosenblatt Perceptron learning algorithm and employs a one-vs-all approach, where each neuron is responsible for recognizing a single class.

The network is composed of three independent perceptrons, each trained using supervised learning with adjustable learning rates and iterative generations. During training, the algorithm tracks the best-performing weights for each neuron and allows interactive testing with unseen input data after training.

The implementation uses Python with NumPy and Pandas for numerical computation and data handling, and retrieves the Iris dataset directly from the UCI Machine Learning Repository.

Features

• One-vs-all multiclass classification using single-layer perceptrons
• Manual implementation of weight update and activation function
• Tracking of best weights and generations per neuron
• Interactive testing with user-provided input data
• Experimental analysis of learning rate influence on convergence

References

1. Iris Data Set – UCI Machine Learning Repository
2. Rosenblatt, F. The Perceptron: A Probabilistic Model for Information Storage and Organization in the Brain, Psychological Review, 1958.

@TalesNogueira