Lung cancer is considered as critical cancers globally, accurate classification of lung nodules plays a major role in early prediction and planning for treatment. In this paper, a hybrid deep learning model that fuses EfficientNet and Autoencoder for multiclass classification of lung nodules has been proposed. Three classes focused are benign, malignant, and metastatic malignant representing nodules with secondary cancerous growths. Feature extraction is performed by utilizing EfficientNet architecture considering its scalability and efficiency. Autoencoder architecture is utilized for dimensionality reduction by capturing important features required for classification in a compressed latent space. The proposed hybrid model is evaluated using LIDC-IDRI dataset containing annotated images of lung CT scan. Experimental results reveal that the performance of the proposed model in terms of classification accuracy and other performance metrics is better compared to existing methods. Also, the proposed hybrid model leads to better generalization during learning while handling high-dimensional data due to its efficiency in noise reduction and feature compression.