Energi listrik merupakan kebutuhan pokok dalam kehidupan modern yang penggunaannya terus meningkat, namun pencurian listrik atau non-technical loss masih menjadi permasalahan serius karena menimbulkan kerugian finansial dan menurunkan kualitas distribusi energi. Penelitian ini bertujuan mendeteksi anomali konsumsi listrik rumah tangga sebagai indikasi pencurian listrik menggunakan metode Gaussian Mixture Model (GMM). Dataset yang digunakan terdiri dari sekitar 17.000 baris data konsumsi listrik rumah tangga dengan variabel energi (Global_active_power, Voltage, Global_intensity, Sub_metering) serta faktor lingkungan (Temperature, Precipitation, CloudCover, Snowfall, Wind). Tahapan penelitian meliputi pra-pemrosesan data (konversi waktu, imputasi nilai hilang, normalisasi), pemodelan dengan GMM menggunakan algoritma Expectation-Maximization, perhitungan log-likelihood, penentuan threshold berbasis persentil, dan deteksi anomali. Hasil penelitian menunjukkan bahwa GMM mampu memodelkan distribusi data konsumsi listrik dengan baik, di mana sekitar 5% data dengan log-likelihood terendah terdeteksi sebagai anomali. Anomali ini diinterpretasikan sebagai pola konsumsi listrik yang tidak wajar dan berpotensi mencerminkan praktik pencurian listrik. Penelitian ini diharapkan dapat berkontribusi dalam pengembangan sistem deteksi dini pencurian listrik berbasis data mining dan machine learning.

Electric energy is one of the essential needs in modern life, with household consumption increasing significantly every year. However, electricity theft or non-technical loss remains a serious problem, causing financial losses and reducing the reliability of energy distribution. This research aims to detect anomalies in household electricity consumption as an indication of electricity theft using the Gaussian Mixture Model (GMM) method. The dataset used consists of approximately 17,000 records of household electricity consumption with Global_intensity, energy-related variables (Global_active_power, Voltage, Sub_metering) and environmental variables (Temperature, Precipitation, CloudCover, Snowfall, Wind). The research stages include data preprocessing (time conversion, missing value imputation, normalization), modeling with GMM using the Expectation-Maximization algorithm, log-likelihood calculation, threshold determination based on percentile, and anomaly detection. The results show that GMM can model the distribution of electricity consumption data effectively, with around 5% of data points having the lowest log-likelihood identified as anomalies. These anomalies are interpreted as irregular electricity usage patterns that may indicate electricity theft. This research contributes to the development of early detection systems for electricity theft using data mining and machine learning approaches.