Machine learning-assisted characterization of 2D materials

In this thesis study, the aim is to analyze the directional and velocity-based displacement behavior on the surface using sequential images obtained from Atomic Force Microscopy (AFM). For this purpose, both classical machine learning algorithms and deep learning approaches were utilized to perform directional classification and drift velocity estimation. In the first stage, displacements in microns along the X and Y axes were used as input variables for directional classification. For this classification task, Random Forest and Support Vector Machine (SVM) algorithms were employed. The Random Forest model achieved a high test accuracy of 98.6%, while the SVM model achieved 94.6%. Moreover, the Random Forest algorithm demonstrated more balanced and robust F1-score results across all classes. In the second stage, drift velocities were calculated based on the time differences and coordinate changes between successive image pairs, and this data was modeled using a Long Short-Term Memory (LSTM) network, which is suitable for sequential data analysis. The LSTM architecture consisted of two layers with 64 and 32 neurons respectively, both using the tanh activation function. The output layer was designed as a single-neuron dense structure. The model was trained using varying numbers of training pairs and evaluated systematically across each configuration, with training conducted for 50 epochs per case. The prediction performance of the model was evaluated using metrics such as Mean Squared Error (MSE), Coefficient of Determination (R²), Root Mean Squared Error (RMSE), and Mean Absolute Error (MAE). Results obtained from the test data revealed that the model was able to make predictions with low error rates, while maintaining high R² scores. These findings confirm that the LSTM model can successfully predict drift velocity when working with time-dependent microscopic data.