Anomaly Detection in Battery Devices: Identifying Unusual Battery Voltage Patterns
(2025) In Master's Thesis in Mathematical Sciences FMSM01 20251Mathematical Statistics
- Abstract
- In modern security systems, the widespread deployment of battery-powered IoT devices creates a critical need for accurate battery replacement strategies. Premature replacement leads to unnecessary waste, while delayed replacement risks system failure in the case of an incident. This thesis presents the workflow building on limited knowledge of the data in an unsupervised learning setting to a working model for behavioural classification in battery voltage time series. Using a combination of manual feature engineering, active learning and machine learning—specifically, the XGBoost algorithm—battery behaviours were categorized and analysed. The results show that not all deviations from normal behaviour are anomalous or difficult to classify.... (More)
- In modern security systems, the widespread deployment of battery-powered IoT devices creates a critical need for accurate battery replacement strategies. Premature replacement leads to unnecessary waste, while delayed replacement risks system failure in the case of an incident. This thesis presents the workflow building on limited knowledge of the data in an unsupervised learning setting to a working model for behavioural classification in battery voltage time series. Using a combination of manual feature engineering, active learning and machine learning—specifically, the XGBoost algorithm—battery behaviours were categorized and analysed. The results show that not all deviations from normal behaviour are anomalous or difficult to classify. Furthermore, weighting time series with different behaviours differently during prediction model training can improve performance of certain behaviours, although it is ultimately not a requirement for good forecasting. A general result is the useful quality-related insights into the IoT-devices, for example different behaviours' prominence and their difficulty in prediction. (Less)
Please use this url to cite or link to this publication:
http://lup.lub.lu.se/student-papers/record/9206963
- author
- Lisra, Victor LU and Ritzing, Milton LU
- supervisor
- organization
- course
- FMSM01 20251
- year
- 2025
- type
- H2 - Master's Degree (Two Years)
- subject
- keywords
- XGBoost, anomaly detection, classification, time series, unsupervised learning, supervised learning, active learning, machine learning
- publication/series
- Master's Thesis in Mathematical Sciences
- report number
- LUNFMS-3536-2025
- ISSN
- 1404-6342
- other publication id
- 2025:E89
- language
- English
- id
- 9206963
- date added to LUP
- 2025-07-04 10:20:40
- date last changed
- 2025-07-04 10:20:40
@misc{9206963, abstract = {{In modern security systems, the widespread deployment of battery-powered IoT devices creates a critical need for accurate battery replacement strategies. Premature replacement leads to unnecessary waste, while delayed replacement risks system failure in the case of an incident. This thesis presents the workflow building on limited knowledge of the data in an unsupervised learning setting to a working model for behavioural classification in battery voltage time series. Using a combination of manual feature engineering, active learning and machine learning—specifically, the XGBoost algorithm—battery behaviours were categorized and analysed. The results show that not all deviations from normal behaviour are anomalous or difficult to classify. Furthermore, weighting time series with different behaviours differently during prediction model training can improve performance of certain behaviours, although it is ultimately not a requirement for good forecasting. A general result is the useful quality-related insights into the IoT-devices, for example different behaviours' prominence and their difficulty in prediction.}}, author = {{Lisra, Victor and Ritzing, Milton}}, issn = {{1404-6342}}, language = {{eng}}, note = {{Student Paper}}, series = {{Master's Thesis in Mathematical Sciences}}, title = {{Anomaly Detection in Battery Devices: Identifying Unusual Battery Voltage Patterns}}, year = {{2025}}, }