Neural Network for Animal Classification

Project Overview

This project involves building and training a multilayer neural network capable of classifying animals into one of seven types based on their characteristics. The neural network is trained using data from the zoo.txt file, which contains details about 121 animals, each described by 18 attributes. The primary goal is to adapt existing code for neural network training to solve this classification problem.

Objectives

1. Learn to adapt and extend neural network training code for a new classification problem.
2. Implement functions to process data and generate training and test sets.
3. Train a neural network to correctly identify animal types.
4. Evaluate the performance of the trained neural network.

Features

• Neural network with backpropagation for training.
• Conversion of animal data into structured training patterns.
• Support for 7 animal types as outputs.
• Evaluation and accuracy calculation for test datasets.

Dataset Description

The dataset is split into:

• Input Attributes: 16 numerical attributes (e.g., feathers, milk production, etc.).
• Animal Type: The final classification output with 7 possible types (e.g., mammal, reptile).

Each line in zoo.txt describes an animal:

• The first value is the animal's name.
• The next 16 values are numerical inputs.
• The last value is the animal type.

Key Functions

1. build_sets

This function processes the dataset and creates:

• Training Set: First 67 patterns.
• Test Set: Remaining patterns.

Steps:

• Read and parse each line from zoo.txt.
• Convert attributes into binary input patterns.
• Generate output patterns based on animal types.
• Randomize the order of patterns.

2. translate

Converts individual animal data into a structured pattern:

• Input: [name, attributes, type]
• Output: [animal_name, input_pattern, animal_type, output_pattern]

Example: Input: ["aardvark", 1, 0, 0, 1, ..., "mammal"] Output:

['aardvark', [1, 0, 0, 1, 0, ..., 1], 'mammal', [1, 0, 0, 0, 0, 0, 0]]

3. train_zoo

• Trains the neural network for 300 iterations using the training set.

• Calls the iterate function to adjust weights for each training pattern.

4. test_zoo

• Tests the trained network using the test set.

• Outputs predictions and compares them with actual types.

• Calculates and prints the success rate.

• Example output:

The network thinks mongoose is a mammal, it should be a mammal
Success rate: 94.12%

5. retranslate

• Converts neural network outputs into corresponding animal types.

Implementation Steps

1. Analyze and understand the provided code for Boolean functions (AND, OR, XOR).

2. Implement the build_sets and translate functions to prepare training and test datasets.

3. Train the neural network using the train_zoo function.

4. Test the neural network using the test_zoo function and evaluate performance.

Evaluation Metrics

• Accuracy: Percentage of correct classifications.

• Output Comparison: Detailed predictions for each test case.

Example Results

• Example input: Attributes of an animal (binary format).

• Predicted output: ['mammal', 'reptile', ...].

• Success rate: e.g., 94.12%.

Prerequisites

• Python: Ensure Python is installed on your system.

• Required libraries: NumPy (or other dependencies mentioned in the provided code).

How to Run

• Place the zoo.txt and info.txt files in the project directory.

• Run the script to process data and train the network.

• Use the test function to evaluate network performance.

References

Course Material: Artificial Intelligence

Provided by: Mr. Arlindo Silva

Files: zoo.txt, info.txt, and sample neural network training code.

Authors

Aziz Zina and Rafik Baaziz.