Age and Gender Prediction using UTKFace Dataset

This project is a deep learning-based age and gender predictor built using the UTKFace dataset. The model is trained with TensorFlow and Keras and utilizes various data science libraries including Pandas, NumPy, Seaborn, and Matplotlib for preprocessing and visualization.

🔍 Overview

The UTKFace dataset contains over 20,000 face images with annotations of age, gender, and ethnicity. This project focuses on:

• Predicting Age (as a regression problem)
• Predicting Gender (as a binary classification problem)

🛠️ Tech Stack

• Python
• Pandas & NumPy
• Matplotlib & Seaborn
• TensorFlow & Keras
• tqdm
• warnings (for cleaner output)

📁 Dataset

The UTKFace dataset contains cropped and aligned face images with filenames formatted as:

[age][gender][race]_[date&time].jpg

Where:

• age is an integer between 0 and 116
• gender is 0 (male) or 1 (female)

Example: 25_0_2_20170116174525125.jpg.chip.jpg → Age: 25, Gender: Male

📊 Data Preprocessing

• Parsed labels (age and gender) from image filenames
• Resized all images to a fixed dimension (e.g. 200x200)
• Normalized pixel values
• One-hot encoded gender labels (if needed for training)
• Split data into training and testing sets

🧠 Model Architecture

Built using Keras with a Convolutional Neural Network (CNN):

• Input Layer: Resized image (e.g. 200x200x3)
• Conv2D + MaxPooling Layers: Feature extraction
• Flatten + Dense Layers
• Output Layers:
  • One node with linear activation for age regression
  • One node with sigmoid activation for gender classification

🏋️ Training

• Loss Functions:
  • Mean Squared Error (MSE) for Age
  • Binary Crossentropy for Gender
• Optimizer: Adam
• Evaluation Metrics: Accuracy (for gender), MAE/MSE (for age)

📈 Results

• Visualized training and validation loss/accuracy
• Plotted sample predictions vs. ground truth
• Displayed confusion matrix for gender classification