Automatic Feature Selection using Ant Colony Optimization (ACO)

This script demonstrates the application of Ant Colony Optimization (ACO) for feature selection in a machine learning pipeline.

Problem Description:

• The script addresses a classification problem aimed at distinguishing between normal and defective wood samples.

• The dataset contains images of both normal and defective wood samples.

• The task involves extracting features from the images and training a classification model to differentiate between the two classes.

  Note that for better accuracy you need to replace the parameters.

1. Data Preprocessing:

   • Images of different tree species are loaded and features are extracted using GLCM (Gray Level Co-occurrence Matrix).
   • The data is split into training, validation, and testing sets.

2. Feature Selection with ACO:

   • ACO algorithm is applied to select an optimal subset of features.
   • The algorithm iteratively selects features based on pheromone levels and heuristic information.
   • Parameters like the number of ants, alpha, beta, and evaporation rate are set to guide the search.

3. Model Training and Evaluation:

   • Support Vector Machine (SVM) classifier is trained using the selected features.
   • Performance is evaluated on the training, validation, and testing sets.
   • Accuracy and classification reports are generated for each set.

Code Explanation:

Data Preprocessing:

• The script loads images of different tree species and extracts features using GLCM.
• Images are resized to a standard size (SIZE = 200) before feature extraction.
• Features extracted include Energy, Dissimilarity, Homogeneity, and Contrast.

Feature Selection with ACO:

• ACO class is defined to perform feature selection.
• The ACO algorithm iteratively selects features based on pheromone levels and heuristic information.
• Parameters like num_ants, alpha, beta, evaporation_rate, and q0 are set to control the behavior of the algorithm.

Model Training and Evaluation:

• Support Vector Machine (SVM) classifier is trained using the selected features.
• Performance is evaluated using accuracy metrics and classification reports.
• Separate evaluations are done for training, validation, and testing sets.

Results:

• The script provides insights into the performance of the ACO-based feature selection method.
• It allows comparison of accuracy and classification reports for different sets.
• The feature importance is visualized through a bar plot of SVM coefficients.

Further Improvements:

• Hyperparameters of ACO and SVM can be fine-tuned for better performance.
• Other feature selection techniques like Genetic Algorithms or Recursive Feature Elimination can be explored.