Utilizing machine learning ensembles for effective electricity theft detection
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
Electricity theft presents significant challenges globally, with traditional detection methods often lagging behind sophisticated techniques. A misuse of authority can have several detrimental effects. These include rising energy consumption, strain on the infrastructure that supplies it, falling power company profits, and risks to public safety such as electrical shocks and fires caused by using electricity. The proposed model used an ensemble method involving voting and stacking methods to train a challenging imbalanced dataset of electricity theft. The ensemble method used logistic regression and random forest models with ADASYN (adaptive synthetic sampling) to achieve the best results. The dataset comprised 1034 customer records (2014–2016), exhibiting marked class imbalance that was corrected to equal class representation using ADASYN. On the ADASYN-balanced data, the stacking model (logistic regression + random forest) delivered class-wise precision/recall/F1 of 0.95/0.94/0.94 for “theft” and 0.94/0.95/0.94 for “non-theft,” with overall accuracy of 0.94. Discrimination performance was strong (ROC-AUC ≈ 0.94), surpassing the voting ensemble (AUC ≈ 0.93) when both were trained on balanced data. Confusion-matrix and metric profiles further show stacking on balanced data outperformed all imbalanced settings and the voting baseline. Experimental results showed that stacking with the combination of logistic regression and random forest achieved the best results from benchmarks of 94% accuracy, recall, and F1-score. These findings indicate a robust, lightweight approach for electricity theft detection that improves minority-class detection without sacrificing overall accuracy.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it