Predictive Vacancy Rate Forecasting Model

Develop a machine learning pipeline using scikit-learn and pandas that predicts property vacancy rates for multi-unit residential complexes. The model should incorporate historical occupancy data, local economic indicators, seasonal trends, and neighborhood development metrics. Create a Flask API endpoint that allows real-time predictions with confidence intervals, and implement cross-validation techniques to ensure model robustness. Include feature importance visualization and a mechanism for periodic model retraining.