Accurate occupancy estimation in transportation is essential for efficient fleet management, optimizing service quality, and ensuring smooth operations. The challenge arises when some trips are equipped with sensors while others operate without them, leading to incomplete datasets. Traditional and deep learning-based imputation methods rely on sensor-equipped trips to estimate occupancy for sensorless trips, but their accuracy suffers when sensor-equipped and sensorless trips are assigned randomly. To address this, a Deep Reinforcement Learning (DRL) approach is introduced to optimize the allocation of sensor-equipped and sensor-less trips within transportation timetables. By dynamically reorganizing trip assignments, this method significantly enhances the accuracy and reliability of imputation models. Results show that the DRL-based allocation approach reduces Mean Squared Error by 10% after optimization. This framework provides a practical and robust solution for managing incomplete datasets, effectively bridging the gap between theoretical imputation models and real-world transportation systems.