Health state prediction is a key technology to ensure the safe and stable operation of the Inertial Navigation System (INS). Aiming at the characteristics of an INS with limited detection times and aperiodic test interval, a method for health state prediction of INS based on aperiodic data is proposed. Firstly, the Metropolis-Hastings algorithm is used to combine the historical detection data of the inertial navigation system of different equipment in the same batch, periodize the detection data of the inertial navigation system of this equipment. Secondly, the principal component analysis method is utilized to extract high-dimensional features to reduce the redundancy and correlation of data. Finally, the health state prediction model of INS is established by using multi-classification support vector machine to predict the health state of INS. The effectiveness and practicability of the proposed method is verified by the historical calibration data of an INS.