🐧 Penguin Species Classifier

An interactive machine learning web application that predicts the species of a penguin based on its physical characteristics. The app leverages a Random Forest Classifier and provides real-time predictions through a user-friendly interface.

🔗 Live App: https://penguins-species-classifier.streamlit.app

Overview

This project demonstrates an end-to-end machine learning workflow integrated into a web application. It combines data preprocessing, model training, and inference within a single pipeline, enabling users to input penguin attributes and obtain species predictions instantly. The application is built using Streamlit for the frontend and Scikit-learn for model development.

Dataset

The dataset used in this project is the cleaned version of the Palmer Penguins dataset, containing observations of penguins from three species:

• Adelie
• Chinstrap
• Gentoo

Dataset Source: Penguins Dataset

Features used:

• Island
• Bill length (mm)
• Bill depth (mm)
• Flipper length (mm)
• Body mass (g)
• Sex

Target variable:

• Species

The data is dynamically loaded from an external source within the application.

Tech Stack

• Frontend: Streamlit
• Backend / ML: Scikit-learn
• Data Handling: Pandas, NumPy

Dependencies:

• streamlit
• numpy
• pandas
• scikit-learn

Machine Learning Model

The application uses a Random Forest Classifier, an ensemble learning method that builds multiple decision trees and aggregates their predictions.

Key characteristics:

• Handles both numerical and categorical features
• Robust to overfitting compared to single decision trees
• Provides class probability estimates

The model is trained dynamically within the app during runtime using the full dataset.

Project Workflow & Pipeline

The application follows a structured machine learning pipeline:

1. Data Loading

• Dataset is fetched from an external CSV source
• Loaded into a Pandas DataFrame
• Features (X_raw) and target (y_raw) are separated

2. User Input Collection

• Inputs are captured via sidebar widgets:
  • Dropdowns for categorical variables
  • Sliders for numerical variables
• A single-row DataFrame is created for user input

3. Data Integration

• User input is concatenated with the original dataset
• Ensures consistent encoding across both training and inference data

4. Feature Engineering

• Categorical variables (island, sex) are encoded using one-hot encoding
• This transformation is applied uniformly to both training data and input sample

5. Target Encoding

• Species labels are mapped to numerical values:
  • Adelie → 0
  • Chinstrap → 1
  • Gentoo → 2

6. Model Training

• A Random Forest Classifier is initialized and trained on the processed dataset
• Training occurs at runtime using all available data

7. Prediction & Inference

• The trained model predicts:
  • Class label (species)
  • Class probabilities for each species
• Output is formatted into a structured DataFrame

8. Output Visualization

• Prediction probabilities are displayed using progress bars
• Final predicted species is highlighted prominently

Application Features

• Interactive input controls for real-time predictions
• Visualization of dataset relationships via scatter plots
• Transparent data exploration using expandable sections
• Probability-based prediction display for interpretability

Key Design Decisions

1. Dynamic Training

The model is trained during each run instead of using a pre-trained model.
This ensures:

• Simplicity of deployment
• No dependency on serialized model files

2. Unified Encoding Strategy

Combining input data with the original dataset before encoding guarantees:

• Consistent feature alignment
• Avoidance of missing dummy variables

3. Lightweight Architecture

The application avoids complex pipelines or external APIs, making it:

• Easy to deploy
• Easy to understand
• Suitable for demonstration and learning purposes

Conclusion

This project showcases the practical integration of machine learning into an interactive web application. It highlights the complete lifecycle, from data preprocessing to real-time inference, within a clean and accessible interface.

The application serves as a strong demonstration of:

• Applied machine learning.
• Data preprocessing techniques.
• Model deployment using Streamlit.