Energy Efficiency Prediction Using R

This project analyzes the Energy Efficiency dataset from the UCI Machine Learning Repository. The dataset is used to predict Heating Load (Y1) and Cooling Load (Y2) based on various building design parameters. The analysis includes correlation studies, model building (Linear Regression and Random Forest), and feature importance evaluation.

---

Dataset Overview

The dataset consists of 768 observations with eight independent variables (X1-X8) and two target variables (Y1 and Y2). All variables are numeric, and the dataset contains no missing values.

Variables:

Variable	Type	Description
X1: Relative Compactness	Numeric (Continuous)	Compactness of the building
X2: Surface Area	Numeric (Continuous)	Total surface area
X3: Wall Area	Numeric (Continuous)	Area of external walls
X4: Roof Area	Numeric (Continuous)	Area of the roof
X5: Overall Height	Numeric (Continuous)	Height of the building
X6: Orientation	Numeric (Categorical)	Cardinal orientation
X7: Glazing Area	Numeric (Continuous)	Window area
X8: Glazing Area Distribution	Numeric (Categorical)	Window distribution
Y1: Heating Load	Numeric (Continuous)	Energy required for heating
Y2: Cooling Load	Numeric (Continuous)	Energy required for cooling

---

Key Features of the Analysis

1. Data Cleaning and Preparation:

   • The dataset is pre-cleaned with no missing values.
   • Numerical normalization was performed for consistent scaling.

2. Correlation Analysis:

   • A Spearman correlation matrix revealed significant relationships:
     • Strong negative correlation: X1 with X2 (-1.00) and X4 (-0.87).
     • Strong positive correlation: X4 and X5 with Y1/Y2 (~0.86).
     • Negligible correlation: X6 with all other variables.

3. Modeling:

   • Models used:
     • Linear Regression
     • Random Forest (500 trees)
   • Dataset split: 70% training, 30% validation.
   • Cross-validation with 10 folds ensures robust evaluation.

4. Results:

   • Random Forest outperformed Linear Regression with lower Mean Squared Errors (MSE) and higher R² values.
   • Feature importance showed that:
     • X2 (Surface Area) is crucial for Y1 (Heating Load).
     • X4 (Roof Area) is the most significant for Y2 (Cooling Load).

5. Visualization:

   • Prediction plots for both models highlight the alignment between actual and predicted values.
   • Feature importance plots identify the top predictors for energy loads.

---

How to Run the Code

Prerequisites:

• Install required R libraries: caret, randomForest, ggplot2, dplyr.

Steps:

1. Load the Dataset:

   • Use ENB2012_data.xlsx for building design parameters and energy loads.
   • Ensure the file is placed in the same directory as the script.

2. Run the Script:

   • Execute the script ICT515-final-exam.R in RStudio or an R environment.
   • The script performs the entire analysis pipeline: loading data, preprocessing, correlation analysis, model training, and evaluation.

3. Outputs:

   • Correlation matrix and significance tests.
   • Prediction plots for both Y1 and Y2.
   • Feature importance plots for both models.
   • Model performance metrics (MSE and R²).

---

Interpretation of Results

1. Correlation Analysis:

   • Variables X2 (Surface Area) and X4 (Roof Area) significantly impact energy loads, correlating positively with both targets.

2. Model Performance:

   • Random Forest captures non-linear relationships and outperforms Linear Regression:
     • Lower MSE.
     • R² values closer to 1.

3. Feature Importance:

   • Emphasizes the role of Surface Area (X2) and Roof Area (X4) in determining heating and cooling requirements.

---

References

• Dataset: Energy Efficiency Data Set, UCI Machine Learning Repository.

---

Author

Kiran Ojha