Neighbor-Based Optimized Logistic Regression Machine Learning Model For Electric Vehicle Occupancy Detection

Machine learning-based solutions for occupancy detection ubiquitously use multilayer Long Short-Term Memory (LSTM) models to achieve high accuracy with large time series data. This paper presents a streamlined and optimized single-layer logistic regression machine learning model that predicts the occupancy of an Electric Vehicle (EV) charging station given the occupancy of neighboring stations. In addition to this new neighbor-based approach, the model was optimized for the time of day. Trained on data from 57 EV charging stations around the University of California San Diego campus, the model achieved an 88.43% average accuracy and 92.23% maximum accuracy in predicting occupancy, outperforming a persistence model benchmark.