As lung cancer has emerged as the top contributor to cancer-related fatalities, efficient and precise diagnostic methods are essential for efficient diagnosis. This research introduces a novel CNN architecture GoogLeNet with Adaptive Layers (GoogLeNet-AL) for lung cancer detection. The GoogLeNet-AL architecture integrates innovative features such as squeeze-and-excitation blocks, dilated convolutions, depthwise separable convolutions, group convolutions, non-local blocks, octave convolutions, inverted Residuals, and ghost convolutions in the inception layers to boost the potential of GoogLeNet-AL to capture multi-scale features efficiently. The GoogLeNet-AL model has been implemented in the PyTorch 1.8.1 platform and trained using publicly accessible datasets IQ-OTH/NCCD and Chest CT-Scan for comprehensive performance evaluation. Additionally, we employ data augmentation, stratified sampling, and fairness-aware training to enhance robustness and mitigate biases. The experimental assessment demonstrate that the GoogLeNet-AL method achieves an accuracy of 98.74 %, an F1-score of 98.96 %, and a precision of 99.74 % in lung cancer detection and also demonstrates its superior performance by outperforming traditional GoogLeNet and other baseline models. Overall, the proposed architecture enhanced the detection and categorization of lung nodules by reducing false positives and negatives, thus offering a valuable tool for combating lung cancer.