Lung cancer screening is a challenging issue. To address the poor classification capacity of single volatile organic compounds (VOCs), we analyzed detection methods for formaldehyde (CH2O) and acetaldehyde (C2H4O), and integrated approaches from previous studies to detect other VOCs, collectively constructing a multi-gas characteristic classification model for lung cancer. Detection of CH2O and C2H4O involved using wavelength modulation spectroscopy (WMS) to detect expired gas. We utilized Improved Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (ICEEMDAN) model and Grey Wolf Optimizer-Bidirectional Long ShortTerm Memory (GWO-BiLSTM) model to denoise signals and invert concentrations, achieving improved sensor performance. The Signal-to-Noise Ratio (SNR) of the system was improved by 26.34 dB, and Root Mean Square Error (RMSE) of CH2O and C2H4O were 0.691 ppb and 0.451 ppb, respectively. The classification model based on random forest (RF) demonstrated a sensitivity of 83.3 % and a specificity of 87.5 %, indicating that this expired gas screening method has the potential to become one of the early screening approaches for lung cancer.