Delaney Solubility Prediction with Molecular Descriptors

This project explores the Delaney solubility dataset using molecular descriptors to predict solubility (logS). We implemented a Linear Regression model as our main approach and compared it with a Random Forest model in the appendix.

---

Linear Regression Model

1. Loading and Preparing Data

• The dataset was loaded from a public repository.
• We created the variable X to store the features used for predicting the target variable.
• The target variable Y contains the solubility values (logS).
• Data was split into training and test sets, with 80% for training and 20% for testing.

2. Model

• A Linear Regression model was created using Scikit-learn.
• The model was trained on the training data and then used to predict values for the test set.

3. Model Evaluation

• Predictions were evaluated using different regression metrics, such as R², MSE and RMSE, to assess the performance of the model.
• These metrics help quantify how well the model captures the variability of the target variable.

4. Data Visualization

• We used Matplotlib to plot the predicted vs. actual logS values.
• Scatter plots provide a visual understanding of how well the model performs and help identify trends and potential deviations.

---

Appendix: Random Forest

• As an additional experiment, we implemented a Random Forest model using Scikit-learn.
• The same train-test split and evaluation metrics were applied to compare its performance with the Linear Regression model.
• According to the metrics, both models performed similarly.
• Since Linear Regression is simpler and more interpretable, it was concluded to be the most suitable model for this problem.

---

Conclusion:
For the Delaney solubility dataset, a simple Linear Regression model is sufficient to capture the underlying trends in the data. Random Forest did not provide a significant improvement in performance, making the linear model the preferred choice for its simplicity and interpretability.