The reliability of switchgear assets is critical for ensuring continuous power delivery in electrical distribution networks. However, aging equipment and recurring defects often lead to costly interruptions if failures are not detected early. This paper presents a data-driven predictive maintenance framework for switchgears based on logistic regression, emphasizing robust preprocessing and probability calibration to achieve reliable defect forecasting. Historical defect records from Tenaga Nasional Berhad (TNB) in Kelantan, covering the period 2020–2024, were consolidated into a defect matrix representing 42 switchgears across 17 inspection cycles. The modeling task was formulated as a binary classification problem to estimate the probability of defect occurrence in the final cycle (C17) and to forecast a hypothetical next cycle (C18). Initial experiments achieved an accuracy of 92.86%, misclassifying only three units. A Leave-One-Cycle-Out (LOCO) analysis revealed that excluding Cycle 15 improved model robustness, increasing accuracy to 95.24% and F1-score to 0.889. Further probability calibration using ROC and calibration curves identified an optimal threshold of 0.332, ensuring both accurate binary decisions and reliable probabilistic predictions. The findings demonstrate that even a simple model such as logistic regression, when supported by rigorous data preparation and calibration, can provide actionable insights for predictive maintenance scheduling. This study highlights the importance of probability calibration and threshold tuning in industrial applications, offering TNB a cost-effective and interpretable predictive maintenance solution for enhancing asset reliability.