AI-BASED HOUSEHOLD ENERGY
CONSUMPTION PREDICTION
Team Leader: Ebinezer M [ 913022104042 ]
Team Members:
Jaya Akash R [ 913022104037 ]
Moorthy S [ 913022104040 ]
Kabilan V [ 913022104045 ]
CONTENT
ABSTRACT
PROBLEM STATEMENT
OBJECTIVE
DATA COLLECTION AND PREPARATION
PROPOSED SOLUTION (METHODOLOGY)
MODEL PERFORMANCE EVALUATION
SCREENSHOTS / DEMONSTRATION
(VIDEO)
FUTURE SCOPE
CONCLUSION
PROBLEM STATEMENT
Rising household energy demand
impacts sustainability and costs.
Predictive analysis can optimize usage
and suggest savings.
OBJECTIVE
Forecast future household energy consumption.
Understand which variables affect energy usage
the most.
Develop a reliable ML model for prediction.
Visualize trends to assist in decision-making.
Promote energy-saving behavior through AI
insights.
DATASET OVERVIEW
DATASET OVERVIEW:
The dataset contains electric power consumption
data for a single household.
Source: UCI Machine Learning Repository.
It includes features like global active/reactive power,
voltage, intensity, and sub-metered energy usage.
Time range: December 2006 to November 2010
(almost 4 years).
Over 2 million rows of minute-interval data.
DATA PREPROCESSING:
Electricity consumption in homes often leads to
unnecessary wastage.
Households lack forecasting tools to manage daily
energy usage.
High peak hour usage stresses energy infrastructure.
No personalized insights for energy reduction are
commonly available.
A predictive solution is needed to optimize household
electricity usage.
METHODOLOGY
APPROACH:
Data cleaning and exploratory data analysis (EDA).
Feature selection and correlation analysis.
Training models like Linear Regression and Random
Forest.
Splitting data into training and testing sets.
Evaluation using regression metrics.
ALGORITHMS USED:
Linear Regression: For basic trend prediction.
Random Forest Regressor: For improved accuracy
using ensemble learning.
Decision Tree Regressor (optional): For simple rule-
based predictions.
EVALUATION METRICS:
Mean Absolute Error (MAE).
Root Mean Squared Error (RMSE).
R² Score (Coefficient of Determination).
Graphical comparison of predicted vs actual values.
Feature importance scores for interpretation.
RESULTS
MODEL PERFORMANCE:
Linear Regression: R² ~ 0.56; MAE ~ 0.3 kWh.
Random Forest: R² ~ 0.84; MAE ~ 0.15 kWh.
RMSE lower for Random Forest indicating better
predictions.
GRAPHS/VISUALIZATIONS:
Line plots of actual vs predicted energy consumption.
Feature importance bar graph (Random Forest).
Heatmap showing feature correlations.
Daily and hourly energy consumption trends.
Histogram of energy usage distribution.
SCREENSHOTS
LINE GRAPH OF TRAINING VS VALIDATION LOSS
SCREENSHOTS
SCATTER PLOT OF ACTUAL VS PREDICTED ENERGY
USAGE
SCREENSHOTS
HEATMAP SHOWING CORRELATION BETWEEN
FEATURES AND TARGET
SCREENSHOTS
PAIRPLOT SHOWING RELATIONSHIPS BETWEEN ALL
NUMERICAL VARIABLES
DISCUSSION
INSIGHTS:
Sub-metering features contribute significantly to
prediction.
Peak usage times align with daily routines.
Voltage fluctuations slightly influence overall
consumption.
Seasonality and weekends show consumption variations.
Predictive models can reliably forecast short-term usage.
CHALLENGES FACED:
Handling missing values due to incomplete records.
Scaling and normalizing diverse units (watts, volts).
High granularity (minute-wise data) increased
computation.
Choosing the right features without overfitting.
Interpretability trade-offs with complex models.
SOLUTION IMPACT
SUSTAINABILITY IMPACT:
Encourages responsible energy consumption.
Reduces energy waste and supports grid load
balancing.
Facilitates integration with renewable energy
planning.
Educates households through consumption
patterns.
Aligns with SDGs like affordable clean energy
and climate action.
PRACTICAL IMPLEMENTATION:
Integrate with smart meters for real-time
prediction.
Offer alerts to users during high energy usage.
Recommend device-specific energy-saving tips.
Assist utility companies in forecasting demand.
Embed in smart home assistants or IoT
dashboards.
CONCLUSION
Successfully predicted household energy usage
using ML.
Random Forest showed the best performance
among tested models.
Visual insights help users and energy providers
alike.
FUTURE WORK:
Incorporate weather and occupancy data for higher
accuracy.
Extend the model for multi-household or community
data.
Real-time deployment with streaming inputs.
Use deep learning for sequential pattern detection.
Explore classification (e.g., normal vs abnormal usage
alerts).
REFERENCES
UCI Machine Learning Repository – Individual household
electric power consumption dataset.
Python libraries: Pandas, Matplotlib, Scikit-learn, Seaborn.
Research articles on energy prediction using ML.
Scikit-learn documentation: Incorporate weather and
occupancy data for higher accuracy
Dataset URL :
https://www.kaggle.com/datasets/samxsam/household-
energy-consumption
