Module 4 workflow tool
This page turns the concrete strength notebook into a step-by-step workflow. It explains the regression problem, shows how the data moves through preprocessing and modeling, and lets you preview evaluation and testing before opening Colab.
Workflow stages
Module 4 is a supervised machine-learning workflow. We give the model many past concrete mixtures together with their real measured strength, and the model learns a pattern that can be used later on a new mix design.
In this module, machine learning means using past data to learn a relationship between concrete ingredients, curing age, and final compressive strength.
MPaThis is a regression problem, not classification. The model predicts a number, not a category.
We train the model parameters. In linear regression these are the coefficients for each feature. In random forest or XGBoost, these are the learned split rules inside many decision trees.
After training, we have a fitted model plus evaluation results such as MAE, RMSE, and R2 that tell us how well it works on unseen data.
We give the trained model a new concrete mix with the same input columns. The model uses what it learned during training to estimate the strength of that new mix.
1
Concrete recipes plus real measured strength values.
2
The model learns patterns that connect inputs to strength.
3
We keep the learned model after training is finished.
4
A new mix goes in and the model returns predicted MPa.
1
Read the CSV and inspect the columns, summary statistics, and correlations.
2
Split the target from the predictors and create train and test sets.
3
Fit baseline and stronger regressors such as linear regression and random forest.
4
Compare MAE, RMSE, and R2 to see which model generalizes better.
5
Enter a new recipe and preview how the workflow produces a strength estimate.
Data setup
This step mirrors the early notebook cells: load the CSV, inspect the data, separate features from the target, and choose how the train/test split and preprocessing should work.
Concrete_Data.csv into a dataframe.| Cement | Slag | Fly Ash | Water | Superplasticizer | Age | Strength |
|---|---|---|---|---|---|---|
| 540.0 | 0.0 | 0.0 | 162.0 | 2.5 | 28 | 79.99 |
| 540.0 | 0.0 | 0.0 | 162.0 | 2.5 | 28 | 61.89 |
| 332.5 | 142.5 | 0.0 | 228.0 | 0.0 | 270 | 40.27 |
| 198.6 | 132.4 | 0.0 | 192.0 | 0.0 | 360 | 44.30 |
Cement, slag, fly ash, water, superplasticizer, coarse aggregate, fine aggregate, and age.
Concrete compressive strength(MPa, megapascals)
Training rows teach the model. Test rows stay unseen until evaluation.
Start with pandas, plotting libraries, and the scikit-learn tools used throughout the workflow.
These cells load the CSV, preview rows, and inspect summary statistics and correlations.
This is where the workflow creates the training and testing partitions.
Scaling is shown after the split so the notebook avoids data leakage.
Model training
Training means the model looks at many known examples and adjusts its internal parameters so its predictions get closer to the real strength values. The notebook first fits a simple baseline, then a stronger tree-based model, and optionally a tuned version.
A baseline model that learns a weighted linear relationship between the inputs and strength.
An ensemble of decision trees that usually performs better on structured data with nonlinear patterns.
An optional boosted-tree model that often performs very well, but may require extra installation and tuning.
1
The model sees many concrete recipes and their known measured strength.
2
It adjusts coefficients or tree rules to better connect inputs to strength.
3
The model keeps improving until predictions are closer to the known target values.
4
After training, we keep the learned model and use it later for evaluation and prediction.
A fitted model object, prediction errors on the test set, and plots or tables that help compare models.
We do not have a perfect formula for every possible mix. The model only learns from the patterns present in the training data.
Testing on unseen mixes helps us check whether the learned model generalizes instead of only memorizing the training set.
The notebook starts with linear regression so every later result can be compared against a simple baseline.
This block switches to the stronger tree-based model selected above.
RandomizedSearchCV searches over a small parameter space to improve the tree model.
This visual mimics the notebook idea that tree models can rank which variables matter most.
Evaluation
Regression evaluation is different from classification. Instead of accuracy, this module uses MAE, RMSE, and R2 to compare how close predictions are to real measured strength values.
These cells calculate MAE, RMSE, and R2 on the held-out test set.
The notebook compares all trained models in one compact results table.
Testing
Prediction happens after training. We enter a new concrete mix with the same input columns, and the trained model estimates its compressive strength. The browser preview below is a teaching demo, while the real notebook prediction comes from the trained scikit-learn model in Colab.
Choose a mix and click "Preview prediction" to see how the workflow would report model outputs.
This final code block shows how the trained model is used on unseen concrete recipes.