My approach

In this project, I applied PCA, t-SNE, LLE, and MDS to transform high-dimensional rock image data into 2D scatter plots, categorizing images by rock types and incorporating actual visuals for enhanced interpretability. I also compared automated feature extraction with human annotations using Procrustes analysis to evaluate alignment of embeddings reduced to 8 dimensions, preserving over 90% variance.For clustering, I utilized K-Means and Expectation-Maximization algorithms to identify distinct groups within the data, aiding in structured analysis. Additionally, I developed a neural network using TensorFlow and Keras, incorporating multiple hidden layers with ReLU activation and a softmax output layer, optimized with the Adam optimizer. This setup was fine-tuned using a validation dataset, improving accuracy and performance in classifying different rock types, making the model highly effective for practical geological applications.