Facial Expression Classification Using Machine Learning

Overview

This project explored the use of supervised machine learning for facial expression recognition, focusing specifically on detecting whether a person in an image was smiling.

The objective was not to build a production biometric system, but to understand the principles behind:

• image preprocessing
• feature extraction
• model training
• classification accuracy
• evaluation under controlled conditions

---

Core Objectives

The system was designed to:

• identify facial features from image data
• train a classifier to distinguish smiling vs non-smiling expressions
• evaluate model performance on unseen data
• experiment with training parameters and dataset variations

The focus was on understanding the machine learning pipeline end-to-end.

---

Model Training & Classification

Using standard Python ML libraries, the project involved:

• preparing and normalising training images
• defining training and validation splits
• fitting a supervised classification model
• evaluating prediction accuracy
• iterating on model configuration

The system learned to recognise patterns associated with smiling faces rather than relying on rule-based logic.

---

Experimentation & Evaluation

Model performance was evaluated using:

• validation accuracy
• comparison between training and testing results
• analysis of misclassifications
• parameter tuning

This provided insight into how data quality, dataset size, and model complexity affect outcomes.

---

Timeframe & Context

• Duration: ~4 months
• Context: 3rd-year university AI project
• Focus: Supervised learning, image-based classification
• Constraint: Academic experimentation rather than production deployment

---

Skills Demonstrated

This project highlights:

• practical application of supervised learning
• image preprocessing techniques
• model training and evaluation workflows
• understanding of overfitting and validation
• translating theoretical AI concepts into working systems

---

Why This Project Matters

This project represents an early, structured engagement with computer vision and machine learning systems.

It demonstrates the ability to:

• move beyond rule-based logic
• structure datasets for training
• evaluate model performance critically
• reason about prediction reliability

It forms part of a broader academic foundation in AI and simulation modelling.

---