In today’s world, energy-efficient buildings are essential for reducing carbon footprints, lowering operational costs, and promoting sustainable living. Heating, ventilation, and air conditioning (HVAC) systems are among the highest energy consumers in modern buildings. Optimizing how and when these systems are used can significantly improve overall energy efficiency. By using data-driven insights, we can predict inefficiencies and take proactive steps to optimize performance — making machine learning a valuable tool in smart building management.
In this project, we worked on predicting the energy efficiency of buildings using machine learning. The idea was to classify buildings as either “efficient” or “inefficient” based on certain input features, including temperature, humidity, occupancy, square footage, and renewable energy usage.
We began by working with a dataset that contained various features relevant to building energy consumption. The dataset included columns like Temperature, Humidity, SquareFootage, Occupancy, RenewableEnergy, and the target variable, EnergyConsumption. From this, we derived a new binary label, Efficiency, based on the median of the energy consumption. If a building consumed more energy than the median, it was labeled as “inefficient”; otherwise, it was labeled as “efficient.”
After filtering out rows where the HVAC system was off, we retained only those records where HVAC Usage was marked “On”, as those were more relevant for analyzing true energy demand and system behavior. This pre-processing helped ensure our model was trained on consistent operational data.
Model Training and Evaluation
For our model, we used the Random Forest classifier from scikit-learn. This algorithm was selected for its robustness and ability to handle non-linear relationships. We used features like temperature, humidity, occupancy, square footage, and renewable energy to predict the efficiency label.
We split the dataset into training and testing sets (80/20 split) and trained the model. After training, we evaluated it using metrics like accuracy, precision, recall, and F1-score. A confusion matrix was also plotted to visualize how well the model was classifying efficient and inefficient buildings.
The results showed that the model performed quite well — giving us around 80% accuracy in classifying building efficiency. We also tested the model manually by feeding it custom inputs to verify its behavior, and it predicted the output correctly.
Building the Gradio Interface
To make the project more interactive and accessible, we built a simple web-based GUI using Gradio. The interface allowed users to input real-time values for temperature, humidity, square footage, occupancy, and renewable energy contribution. Upon submitting, the model processed the input and displayed whether the building was efficient or inefficient.
This interface made the model easy to use for people without any programming background — allowing quick testing and demonstration of the system’s predictive power. It served as a mini dashboard for energy efficiency checking.
Conclusion
With this project, we successfully demonstrated how machine learning can be applied to building energy data to predict energy efficiency. Through the combination of data cleaning, feature engineering, model training, and UI development, we created a tool that could potentially assist in smarter energy management practices — and even extend into energy-saving automation in smart buildings.