Building a Convolutional Neural Network for Hair Type Classification: A Hands-On Approach

In the Machine Learning Zoomcamp 2024, led by Alexey Grigorev at DataTalksClub, we participants are tasked with building a convolutional neural network (CNN) for classifying hair types. Unlike using pre-trained models, the goal here is to design a model from scratch to handle a dataset of hair images, which will be split into training and test sets. This exercise provides a deep dive into the essential principles of CNNs, including data preparation, model construction, and evaluation.

Dataset and Model Architecture

The dataset for this homework consists of approximately 1,000 images of hair, divided into training and test sets. Each image is of size 200x200x3 (200 pixels by 200 pixels with 3 color channels—RGB). The objective is to design a CNN that will learn from this dataset and predict the hair type. The model construction follows a typical CNN pipeline, beginning with input processing and progressing through various layers.

Key Layers in the Model

  1. Input Layer: The model begins by accepting images of shape (200, 200, 3), representing 200x200 pixel images with 3 RGB channels.

  2. Convolutional Layer: The first convolutional layer utilizes 32 filters with a (3, 3) kernel size, applied to the input image. The ReLU activation function is used to introduce non-linearity, enabling the model to learn complex patterns in the images.

  3. Max Pooling Layer: The next step involves reducing the spatial dimensions of the feature maps produced by the convolutional layer. This is achieved using a max pooling layer with a pool size of (2, 2), which helps reduce computational complexity and retains important features.

  4. Flatten Layer: The feature maps from the max pooling layer are flattened into a one-dimensional vector to be passed to the fully connected layers.

  5. Dense Layer: A fully connected Dense layer with 64 neurons is added. This layer is again activated using the ReLU function, enabling the model to learn more complex relationships in the data.

  6. Output Layer: The final output layer consists of a single neuron with a sigmoid activation function. This activation is ideal for binary classification tasks, as it outputs a probability between 0 and 1, indicating whether the hair type is straight or curly.

Model Parameters

The model’s total number of parameters is a key indicator of its complexity. It’s important to note that the 896 parameters in the convolutional layer represent just part of the model's total complexity. The complete model, including the dense layers, has a total of 20,073,473 parameters. This high number of parameters is indicative of the model's ability to learn intricate features from the data, though it also increases the risk of overfitting if not carefully managed.

Training the Model

The model is trained using the Stochastic Gradient Descent (SGD) optimizer, with a learning rate of 0.002 and momentum set to 0.8. These values help the optimizer navigate the loss surface efficiently and avoid local minima, ensuring that the model converges during training.

The model’s performance is evaluated by tracking both the accuracy and the loss over multiple epochs. The median training accuracy for the model across all epochs is approximately 0.72, indicating that the model is able to learn from the data reasonably well. Furthermore, the standard deviation of the training loss is 0.068, suggesting that the loss varies slightly across epochs but generally decreases over time.

Enhancing Performance with Data Augmentation

Data augmentation plays a crucial role in improving the model’s generalization capability. The model is trained with additional transformations applied to the training data, such as rotation, width and height shifts, zoom, and horizontal flipping. These augmentations artificially expand the dataset, preventing overfitting and allowing the model to generalize better to unseen data.

With the augmented data, the model continues training for an additional 10 epochs. The mean test loss for this extended training phase is found to be 0.56, showing an improvement in generalization performance. The average test accuracy over the last five epochs (epochs 6 through 10) reaches 0.71, demonstrating that the augmented model performs better on the test data.

Model Evaluation and Results

Upon completing the model training, the final evaluation metrics show that the model performs quite well given the complexity of the task. The test accuracy demonstrates an improvement after data augmentation, reflecting the model's ability to generalize better to new, unseen examples.

  • Total Parameters: 20,073,473
  • Median Training Accuracy: 0.72
  • Standard Deviation of Training Loss: 0.068
  • Mean Test Loss After Augmentation: 0.56
  • Average Test Accuracy (last 5 epochs): 0.71

These results demonstrate that the CNN, even when trained from scratch, can achieve reasonable performance in a complex image classification task like hair type classification.

Conclusion

Building a CNN from scratch to classify hair types offers a valuable hands-on experience in understanding the inner workings of deep learning. By following a straightforward architecture and applying techniques such as data augmentation, one can significantly enhance the model's ability to generalize to new data. The process emphasizes the importance of thoughtful model design, careful training, and ongoing evaluation to improve performance in real-world applications.

This project is a great starting point for anyone looking to deepen their understanding of CNNs and their practical applications in image classification tasks.

Comments

Post a Comment

Popular posts from this blog

My midterm project at MLZoomcamp led by Alexey Grigorov for DataTalksClub

 Predicting Patient No-Shows: A Data-Driven Approach Hospital no-shows significantly disrupt healthcare systems, wasting resources and delaying care for those in need. My midterm project for the MLZoomcamp , led by Alexey Grigorev and hosted by DataTalksClub, tackles this challenge using machine learning to predict no-show probabilities for appointments in Brazilian hospitals. Here's how I approached the problem: The Challenge The dataset, sourced from Kaggle, includes over 110,000 appointments and diverse features such as patient demographics, appointment details, and medical history. However, achieving reliable predictions is complex due to: Imbalanced Data : About 80% of appointments were attended, while 20% were no-shows. Dependence on Feature Engineering : Key predictors like patient history (previous/missed appointments) were engineered from the raw data. Bias Mitigation : Socioeconomic factors, such as neighborhood, required careful handling to ensure fairness. The Solution ...
Read more

Logistic Regression: A walkthrough by Alexey Grigorev

Logistic Regression is one of the foundational algorithms for classification tasks, and Alexey Grigorev at DataTalks.Club provides an insightful approach to understanding it in his courses, such as the Machine Learning Zoomcamp. Alexey Grigorev provides a clear and practical explanation of Logistic Regression, helping learners understand its application and limitations in real-world scenarios. In his teaching, Alexey emphasizes the simplicity and interpretability of Logistic Regression. The model predicts the probability that a given input belongs to a particular class, which makes it especially useful in binary classification problems. For example, it’s widely applied in fields like customer churn prediction, fraud detection, and medical diagnoses. Logistic Regression works by modeling the relationship between input features and the probability of a binary outcome using a sigmoid function. This makes the predictions constrained between 0 and 1. Alexey often stresses the importance of ...
Read more

Linear Regression: A Deep Dive with Alexey Grigorev

  Linear Regression: A Deep Dive with Alexey Grigorev Linear regression is a cornerstone of machine learning, and Alexey Grigorev's teachings at Data Talks Club provide a comprehensive and insightful exploration of this fundamental algorithm. Key Concepts Covered by Alexey Grigorev: Simple Linear Regression: Understanding the relationship between a single independent variable and a dependent variable. Multiple Linear Regression: Modeling relationships with multiple independent variables. Assumptions: Exploring the underlying assumptions of linear regression, such as linearity, independence, normality, homoscedasticity, and no multicollinearity. Model Evaluation: Learning how to evaluate the performance of a linear regression model using metrics like R-squared, mean squared error (MSE), and root mean squared error (RMSE). Regularization: Understanding techniques like Ridge and Lasso regression to prevent overfitting and improve model generalization. Feature Engineering: ...
Read more