Inspiration
This project reflects the base for my research on using regression to predict compressive strength.
Previous Work vs Now
Before this day in Open-Source Week: Simply parsed through the dataset and left the code with an incomplete fuzzy clustering + linear regression model. The test R2 score obtained was 0.65. After this day in Open-Source Week: Conducted EDA deliberately with clear end goals. Selected a regressor model based on past experience and existing results. Trained the model complete with KFold cross-validation. The test R2 obtained was 0.88.
What it does
It saves experimentation time for civil engineers by estimating its target based on relevant input features such as cement content, water content, etc.
How we built it
The project is built entirely in Python. It uses the Pandas and NumPy modules for data manipulation, Matplotlib and Seaborn for visualization, and scikit-learn for preprocessing and model training.
Challenges we ran into
Our biggest challenge was to overcome overfitting as well as underfitting by finding the right train-test split to train the model. Another constraint was not being allowed to remove any features from the dataset.
Accomplishments that we're proud of
The model achieves a best R2 of 92, which is great for a small dataset with no feature reduction.
What we learned
Find a model for your data, not the other way around.
What's next for Predicting Compressive Strength with Random Forest Regressor
Improving model scores with more sophisticated ensembles like LightGBM, or ensembles that allow for more model diversity.
Log in or sign up for Devpost to join the conversation.