Quantitative Analysis and Predictive Modeling of Concrete Strength Based on Composition

Project Overview

This project focuses on predicting the compressive strength of concrete, which is a crucial measure in civil engineering. I utilize various machine learning models to understand the relationship between the concrete mixture constituents and its strength.

Goals

• To explore the dataset and understand the factors affecting concrete strength.
• To develop a predictive model with the highest possible accuracy.
• To provide insights and recommendations for concrete mix optimization.

Initial Hypotheses

• There a realtionship between the amount of cement in a cubic yard and concrete strength of a sample.
• There a realtionship between the amount of coarse aggregate in a cubic yard and concrete strength of a sample.
• There a realtionship between the amount of fine aggregate in a cubic yard and concrete strength of a sample.
• There a realtionship between the amount of total weight in a cubic yard and concrete strength of a sample.

Project Plan

1. Acquire

• get the data into pandas
• look at it
• describe, info, head, shape
• understand what your data means
• know what each column is
• know what your target variable is

2. Wrangle

• clean the data
• handle nulls
• handle outliers
• correct datatypes
• univariate analysis (looking at only one variable)
• encode variables -- Preprocessing
• split into train, validate/, test
• scale data (after train/validate/test split) -- Preprocessing
• document how you're changing the data

3. Explore

• use train data
• use unscaled data
• establish relationships using multivariate analysis
• hypothesize
• visualize
• statistize
• summarize
• feature engineering
• use scaled data

4. Model

• use scaled/encoded data
• split into X_variables and y_variables
• X_train, y_train, X_validate, y_validate, X_test, y_test
• build models
• make
• fit (on train)
• use
• evaluate models on train and validate
• pick the best model and evaluate it on test

5. Test

• present results of the best model

Data Dictionary

Feature	Definition
cement	Quantity of cement in the sample mix (yd³)
slag	Quantity of slag in the samplemix (yd³)
ash	Quantity of fly ash in the samplemix (yd³)
water	Quantity of water in the sample mix (yd³)
superplastic	Quantity of superplasticizer in the sample mix (yd³)
coarseagg	Quantity of coarse aggregates in the sample mix (yd³)
fineagg	Quantity of fine aggregates in the sample mix (yd³)
age	Age of concrete at testing (days)
strength	Compressive strength of concrete (psi)
total_lbs_per_yd³	Total weight of all components per cubic yard (yd³)
sample	Identifier for the concrete sample

Steps to Reproduce

1. Download concrete_strength.ipynb, wrgangle.py, and explore.py files in this repository.
2. Obtain the required .csv datased from 'https://www.kaggle.com/datasets/vinayakshanawad/cement-manufacturing-concrete-dataset'
3. Run it

Takeaways

• The Polynomial Regression model significantly outperforms the baseline, explaining up to 47.92% of the variance in concrete strength.¶
• With 'total_lbs_per_yd^3' positively correlated and 'coarseagg' and 'fineagg' negatively correlated with strength, all significant at an alpha below 0.05.

Recommendations

• Stakeholders should prioritize optimizing the total mix composition, particularly focusing on the 'total_lbs_per_yd^3' due to its strong positive impact on strength.
• A deeper analysis of 'coarseagg' and 'fineagg' proportions is advised to enhance the concrete's strength further.

Next Steps

• Explore additional features or transformations that might better capture the relationships in the data, such as interaction terms between different types of aggregates.

Dataset Source: Kaggle - Cement Manufacturing Concrete Dataset