Cement Compressive Strength Prediction Model Results

28-day compressive strength data from cement plants was used to validate a hybrid Transformer-XGBoost prediction model. The model achieved an average R² of 0.94 in 25 Monte Carlo cross-validations, demonstrating high accuracy for small-sample scenarios. The dataset contains the results of this optimization study, authored by Dianyuan Ju and shared in 2026.

Use Cases

• Analyze the performance metrics like R² and RMSE from the Transformer-XGBoost model validation.
• Compare the robustness to noise and handling of feature missingness documented for the XGBoost meta-learner.
• Examine the 28-day compressive strength testing data used for real-world model validation from cement plants.
• Study the attention score weights dynamically allocated by the Transformer's multi-head attention mechanism on physicochemical variables.

Strengths

• Model validation achieved a high average R² of 0.94 across 25 Monte Carlo cross-validations.
• The proposed method demonstrated a 5.62% increase in R² and a 22.33% decrease in RMSE compared to a baseline.
• Results are based on real-world 28-day strength testing data from cement production facilities.

Limitations

• The dataset is very small at 5.5 KB, indicating it contains summary results, not the raw training or test data.
• The scope is limited to the specific TF-XGBoost model optimization study, not a broad collection of cement properties.
• The dataset's utility is confined to reviewing the study's findings rather than enabling new model training.

Provenance

Source

Dianyuan Ju

Collection Method

Real-world 28-day strength testing data from cement plants used for model validation.

Freshness

Last updated in March 2026.