Speed of sound prediction

This machine learning project predicts the speed of sound in electrolytic aqueous solutions.

Directory structure

instructions.md - file with instructions on experiment running.
data - directory with input CSV files.
datasets - directory with preproccesed datasets.
notebooks - directory with Jupyter notebooks.
htmls - directory with all notebooks exported as HTML files.
environment.yml - conda environment file with Python dependencies.

Prerequisities

All dependencies are listed in environment.yml file.

Running the experiment

To run the full experiment, the following notebook order must be maintained. All notebooks are stored in notebooks.

Step 1: data_loading.ipynb

loads CSV files from data/
contains data visualizations

Step 2: features_addition.ipynb

enhances dataframes loaded in Step 1 with ion specific features
scales all the ion specific features
concatenates dataframes for all media into one
saves the full dataframe as dataset.csv into datasets/

Step 3: baseline_models.ipynb

loads dataset.csv created in Step 2
normalizes temperature and molality features
splits data into train, validation and test sets
tests baseline models

Step 4: svm_model.ipynb and rf_model.ipynb

performs Bayesian hyperparameter optimization on SVM, respectively random forest model

Step 5: feature_selection.ipynb

performs forward and backward feature selection
Step 4 can be repeated with new feature sets

Step 6: model_evaluation.ipynb

evaluates optimized SVM and random forest models on three tests
- general test
- inside matrix test
- outside matrix test