Machine Learning Approaches for Aftermarket Demand Forecasting: Tackling Intermittent Time Series Challenges