Monitoring incremental concept and data drift is important for electric load forecasting in smart buildings. This study proposes TPDM-AR (Two-Phase Drift Monitoring and Automatic Retraining), a customizable workflow that combines Adaptive Windowing (ADWIN), Kolmogorov–Smirnov Windowing (KSWIN), and Page–Hinkley (PH) drift detectors with a multilayer perceptron (MLP) regressor. Phase 1 monitors the raw active-power stream for distribution shifts, while Phase 2 confirms predictive degradation by tracking mean absolute error (MAE) using a smoothed MAE trajectory compared with a baseline reference threshold defined from a reference window. Retraining is initiated only when Phase 2 satisfies a consensus-and-persistence confirmation rule on the forecasting-error stream, while Phase 1 alerts on the raw series are reported as complementary signals. The method is validated on controlled synthetic linear and logarithmic drift scenarios and on the UCI Individual Household Electric Power Consumption dataset, reporting detector alert times and the corresponding evolution of forecasting error under the same monitoring setup. • Two-phase drift monitoring that combines early change alerts with performance confirmation. • Retraining is triggered only after Phase 2 consensus-and-persistence confirmation on MAE. • Validated on synthetic drift and real-world electricity-demand data.

Context: Machine learning has become an essential tool for addressing complex problems in information systems, encompassing industrial, commercial, and residential applications. Problem: Machine learning systems without frequent retraining are prone to data and concept drift, compromising predictive accuracy. This issue is particularly critical in scenarios where retraining is infeasible due to high computational costs or data unavailability. Solution: This study evaluates the performance of drift detection methods in discrete time series with controlled changes in mean and standard deviation using synthetic Gaussian signals. IS Theory: The General Systems Theory underpins the study by emphasizing how the interplay between drift detection and adaptive systems contributes to maintaining stability and efficiency in dynamic environments. Method: Experiments were conducted with variations in mean, standard deviation, and both parameters simultaneously in order to obtain qualitative patterns of drift detectors behaviors. The detectors ADWIN, KSWIN, and Page-Hinkley were tested under this scenario. Summary of Results: The findings reveal that ADWIN and Page-Hinkley exhibited greater precision and robustness, while KSWIN showed excessive sensitivity, leading to a high number of false positives. Contributions to the IS Field: This research offers a comprehensive analysis of drift detectors’ performance, specifically in scenarios involving changes in mean and standard deviation, providing useful reference for designing resilient machine learning-based forecasting systems. Impacts on the IS Field: The study advances the development of information systems that can adapt to dynamic data environments characterized by shifts in mean and standard deviation, with direct applications in industrial contexts and energy management.

Ensuring the safety and reliability of photovoltaic (PV) installations requires accurate electrical arc fault detection. This work presents a computational arc fault detection framework that combines fixed-length windowing, Fast Fourier Transform (FFT)-based features, and supervised machine learning classifiers. Data were acquired using an Arc Fault Circuit Interrupter (AFCI) test bench developed based on IEC 63027. Current and voltage signals were partitioned into 200-sample windows, DC-offset corrected, and Hann-windowed signals. Each window generated 204 statistical and spectral attributes used to train and evaluate Random Forest (RF), K-Nearest Neighbors (KNN), Multilayer Perceptron (MLP), and Convolutional Neural Network (CNN) models. Hyperparameters were tuned by grid search with TimeSeriesSplit cross-validation, comparing min–max normalization and Z–Score standardization. On a 15% hold-out test set, CNN with Z–Score achieved F1 = 0.9982 and recall = 0.9975, followed by MLP (F1 = 0.9957) and RF (F1 = 0.9821). Amortized per-window inference latencies were ≈0.0035 ms for RF, ≈0.0016 ms for MLP with Z–Score, and ≈0.032 ms for CNN. These classifier-stage timings indicate computational compatibility with edge-oriented implementation but do not constitute an end-to-end IEC 63027 AFCI compliance assessment. The framework targets integration into PV inverters at Mackenzie Presbyterian University’s solar plant.