🧱 Smart Infrastructure: Predicting Concrete Compressive Strength using Machine Learning

This project uses machine learning to predict the compressive strength of concrete based on its ingredients. The model assists in optimizing mix designs, reducing material waste, and enhancing structural reliability.

📂 Dataset

The dataset used is the Concrete Compressive Strength Dataset from the UCI Machine Learning Repository. It includes 1030 samples with the following features:

• Cement (kg/m³)
• Blast Furnace Slag (kg/m³)
• Fly Ash (kg/m³)
• Water (kg/m³)
• Superplasticizer (kg/m³)
• Coarse Aggregate (kg/m³)
• Fine Aggregate (kg/m³)
• Age (days)
• Target: Concrete Compressive Strength (MPa)

🧠 Model

A simple feedforward neural network (Multilayer Perceptron) is used to learn and predict the target variable.

🔧 Libraries Used

• pandas, numpy
• sklearn for preprocessing and splitting
• tensorflow and keras for building and training the neural network
• matplotlib for visualization

📊 Evaluation

• Loss function: Mean Squared Error (MSE)
• Performance Metric: R² Score

The model shows promising accuracy in learning the non-linear relationship between ingredient proportions and concrete strength.

📈 Results

The model is trained over 100 epochs. Plots of loss curves and predictions versus actual values are included in the code.

📁 Files

• Concrete_Data.xls: Raw dataset
• smart_infra_project.py: Model building, training, and evaluation code
• Smart_Infra_Project.pdf: Project report with background, methodology, and conclusions

🚀 How to Run

1. Clone this repo:

   git clone https://github.com/your-username/smart-infra-concrete-strength.git
   cd smart-infra-concrete-strength
   

2. Install dependencies:

   pip install -r requirements.txt
   

3. Run the script:

   python smart_infra_project.py
   

📁 Project Structure

smart-infra-concrete-strength/
│
├── data/
│   └── Concrete_Data.xls               # Raw dataset
│
├── plots/                              # Saved training/evaluation plots (to be added)
│   ├── loss_curve.png
│   ├── mae_plot.png
│   └── prediction_vs_actual.png
│
├── smart_infra_project.py              # Full code for data preprocessing, training and optimization
├── Smart_Infra_Project.pdf             # Detailed report (includes results and explanation)
├── requirements.txt                    # List of Python dependencies
└── README.md                           # Project overview and instructions

📊 Model Summary

• Architecture: Feedforward Neural Network
• Best configuration: 3 hidden layers [128, 64, 32] with ReLU activation
• Optimizer: SGD with momentum
• Evaluation Metrics:
  • Mean Squared Error (MSE): ~29.93
  • Root Mean Squared Error (RMSE): ~5.47
  • Mean Absolute Error (MAE): ~3.98
  • R² Score: ~0.89

📈 Results Visualization

Training and validation progress:

Training and Validation MSE Loss

Mean Absolute Error over epochs

Predicted vs Actual Strength

🧪 Additional Features

• Learning rate schedules (step decay, cosine annealing)
• Regularization using dropout layers
• Final model saved as concrete_strength_prediction_model.h5
• Optimization utility to suggest mix designs for a target compressive strength

🧠 Potential Enhancements

• Add other ML models for comparison (Random Forest, XGBoost, etc.)
• Integrate with a simple Streamlit or Flask web app
• Use SHAP for better feature importance insights
• Make the optimizer results exportable (CSV)

📄 License

This project is licensed under the MIT License. See the LICENSE file for details.

---

👨‍🔬 Author

• Alok Godara
  Indian Institute of Technology, Jodhpur

---

Feel free to fork, star ⭐, or contribute to this project!