The single-detector spectrometers based on 2D layer van der Waals (vdW) heterojunctions offer advantages in spectral reconstruction due to their high sensitivity, tunable optical properties, and the ability to cover a broad spectral range. There exist two principal algorithms dominating spectrum reconstruction for this kind spectrometer: the Tikhonov regularization method combined with the Least Squares Method (LSM) and neural network-based approaches, particularly Deep Learning (DL). However, both of the algorithms exhibit inherent limitations in spectral reconstruction, which constrain the versatility of computational spectrometers that rely solely on a single algorithm for reconstructing diverse spectral profiles. To overcome this limitation, we introduce an artificial neural network (ANN)-based classification model capable of dynamically selecting the optimal algorithm throughout the reconstruction process. This enables highly accurate spectral reconstruction within the 440-700 nm wavelength range, achieving a spectral resolution of 6 nm. By harnessing the complementary strengths of multiple algorithms, our approach proposes a novel strategy for combining techniques to enhance the precision of spectral reconstructions, laying the groundwork for more sophisticated methods in the future.