Interpreting Lung Cancer Predictions Using CNN-Derived Features and Machine Learning Models

In this paper, the local and global feature importance for lung cancer diagnosis is explained using SHAP (Shapley Additive explanations) and LIME (Local Interpretable Model-agnostic Explanations). Firstly, the images from the IQ-OTH/NCCD Lung Cancer Dataset for three classes - normal, malignant, benign of the dataset are pre-processed and feature extraction is done using various CNN (Convolution Neural Network) architectures, namely - VGG16, ResNet - 50, InceptionV3, MobileNetV2, EfficientNet - B0. The extracted features are used to train models based on various Machine Learning (ML) classifiers. From the outcome of the classifications, it is observed that SVM with VGG-16 based feature extraction offers superior performance with precision, recall and F1 score of 97%, with an average accuracy of 98%. Further, explainable AI frameworks (SHAP, LIME) are used to produce insights into the model's predictions.