Build Accurate Energy Consumption Prediction Model for Utilities

Q: Build Accurate Energy Consumption Prediction Model for Utilities

Amazon machine learning prompt on building an energy consumption forecasting model for utilities, covering data acquisition, EDA, feature engineering, baselines, time-series validation, deployment, monitoring, and retraining.

Q: How do I approach Machine Learning interview questions?

Machine Learning questions require understanding of core concepts and practice. PracHub provides solutions with explanations to help you master machine learning interviews.

Q: What difficulty level is this interview question?

This is a hard difficulty Machine Learning question, commonly asked during Onsite rounds at Amazon.

Q: What role is this question designed for?

This question is commonly asked for Data Scientist candidates at Amazon during technical interviews.

Question

Build an Energy Consumption Prediction Model for Utilities

You need to build and productionize a regression model that predicts daily energy consumption for utility clients. The data is time-indexed and likely depends on weather, seasonality, customer behavior, holidays, tariffs, and meter quality.

Constraints & Assumptions

Predict next-day daily energy consumption at the client, meter, account, or aggregate level; state the chosen granularity.
Use only information available at prediction time.
Include baselines before complex models.
Use validation that respects time ordering and client or meter grouping.
Cover the full path from raw data to production monitoring.

Clarifying Questions to Ask Guidance

What forecast horizon and cadence are needed?
Is the target per meter, per customer segment, feeder, region, or whole utility?
What actions depend on the forecast, and what error metric matters most?
Are weather forecasts available at prediction time?
What latency, interpretability, and reliability requirements apply?

What a Strong Answer Covers Guidance

Problem framing: target, horizon, granularity, success metrics, and business use.
Data acquisition: meter reads, weather observations and forecasts, calendar, holidays, tariffs, outages, customer attributes, and operational data.
Data quality and EDA: missing reads, duplicates, outliers, timezone/DST issues, meter changes, seasonality, trend, autocorrelation, and segment differences.
Feature engineering: lag features, rolling averages, day-of-week, month, holidays, heating/cooling degree days, weather interactions, customer segment, tariffs, and recent anomalies.
Baselines: yesterday, same day last week, seasonal averages, and simple linear or persistence models.
Model selection: regularized regression, tree ensembles, gradient boosting, random forest, hierarchical or panel models, and time-series approaches where appropriate.
Validation: time-based splits, rolling-origin backtests, grouped evaluation by customer/meter, leakage checks, and segment-level metrics.
Evaluation and residual analysis using MAE, RMSE, WAPE, bias, calibration by segment, and error during peak demand or extreme weather.
Productionization: feature pipelines, model registry, batch scoring, data contracts, alerting, rollback, retraining, and monitoring for drift and performance decay.

Follow-up Questions Guidance

Why might MAPE be a poor metric for low-consumption customers?
How would you handle missing weather forecasts?
What drift signal would trigger retraining?
How would you explain the forecast to utility operators?

Build Accurate Energy Consumption Prediction Model for Utilities

Quick Overview

Build Accurate Energy Consumption Prediction Model for Utilities

Build an Energy Consumption Prediction Model for Utilities

Constraints & Assumptions

Clarifying Questions to Ask Guidance

What a Strong Answer Covers Guidance

Follow-up Questions Guidance

Write your answer