[Paper Review] Predicting Smartphone Battery Life based on Comprehensive and Real-time Usage Data

Huoran Li, Xuanzhe Liu+1 more|arXiv (Cornell University)|Jan 12, 2018

Green IT and Sustainability24 references16 citations

TL;DR

This paper proposes a machine learning-based method for predicting smartphone battery life using comprehensive, real-time usage data from 51 users over 21 months. By leveraging fine-grained system, sensor, and app-level traces and addressing data missingness via the concordance index, the model reduces prediction error by 33 minutes on average, significantly improving accuracy over baseline methods.

ABSTRACT

Smartphones and smartphone apps have undergone an explosive growth in the past decade. However, smartphone battery technology hasn't been able to keep pace with the rapid growth of the capacity and the functionality of smartphones and apps. As a result, battery has always been a bottleneck of a user's daily experience of smartphones. An accurate estimation of the remaining battery life could tremendously help the user to schedule their activities and use their smartphones more efficiently. Existing studies on battery life prediction have been primitive due to the lack of real-world smartphone usage data at scale. This paper presents a novel method that uses the state-of-the-art machine learning models for battery life prediction, based on comprehensive and real-time usage traces collected from smartphones. The proposed method is the first that identifies and addresses the severe data missing problem in this context, using a principled statistical metric called the concordance index. The method is evaluated using a dataset collected from 51 users for 21 months, which covers comprehensive and fine-grained smartphone usage traces including system status, sensor indicators, system events, and app status. We find that the remaining battery life of a smartphone can be accurately predicted based on how the user uses the device at the real-time, in the current session, and in history. The machine learning models successfully identify predictive features for battery life and their applicable scenarios.

Motivation & Objective

• To address the critical challenge of smartphone battery life as a bottleneck in daily user experience.
• To overcome the limitations of prior studies that rely on coarse, lab-based, or short-term data.
• To develop a machine learning model that leverages fine-grained, longitudinal smartphone usage traces for accurate battery life prediction.
• To identify and resolve the severe data missing problem—distinguishing between battery death and charging before depletion—using a principled statistical metric.
• To provide a generalizable, system-level prediction framework applicable across diverse smartphone models and usage contexts.

Proposed method

• Collects comprehensive, real-time usage traces from 51 users over 21 months, including system status, sensor readings, system events, and app activity at high temporal granularity.
• Applies state-of-the-art machine learning models, particularly tree-based models, to predict remaining battery life based on current, session-based, and historical usage patterns.
• Uses the concordance index from survival analysis as a key metric to handle censored data—specifically, distinguishing between cases where users charge before battery depletion versus actual battery death.
• Designs predictive features based on discharging history, current and past system status, app usage, and time-of-day patterns.
• Employs a standard machine learning pipeline allowing retraining on new datasets and generalization to other devices like wearables or electric vehicles.
• Validates model performance using metrics including mean absolute error, Kendall’s Tau, and concordance index, comparing against baseline linear regression models.

Experimental results

Research questions

• RQ1 Can fine-grained, real-world smartphone usage data significantly improve the accuracy of remaining battery life prediction compared to coarse or lab-based data?
• RQ2 How does the inclusion of historical usage patterns—especially discharging behavior—impact prediction performance?
• RQ3 To what extent can user-specific usage behaviors enhance the predictive power of battery life models?
• RQ4 How can the common problem of missing data in battery life prediction—where users charge before battery death—be effectively modeled and mitigated?
• RQ5 Can tree-based models outperform linear models in capturing the non-linear relationships between usage patterns and battery drain?

Key findings

• The model reduces the average prediction error by 33 minutes compared to baseline methods, a substantial improvement with practical implications for user task planning.
• Battery discharging history is the most predictive feature, indicating that past usage patterns are more informative than current system status alone.
• User-specific features, such as usage habits and app preferences, provide additional predictive power, supporting personalized battery life estimation.
• System status features are more informative when discharging rates are stable; otherwise, user-specific features dominate, indicating context-dependent feature importance.
• Tree-based models outperform linear regression, confirming that the relationship between usage behavior and battery drain is inherently non-linear.
• The concordance index effectively handles data missingness due to pre-depletion charging, significantly improving model robustness and accuracy in real-world scenarios.