Competition hosted on Analyticsvidhya
Develope a robust and high-performance model utilizing computer vision techniques to classify images as either fraudulent or non-fraudulent within the context of insurance claims. By precisely identifying fraudulent images, insurance companies can evaluate the authenticity of a claim and make well-informed decisions regarding payout.
-
* seaborn * Pandas * Numpy * Matplotlib * imagehash * distance * Image * cv2
-
Model
Trained efficientvit_b3_r224 model on five-fold training data with various augmentations. Ten epochs were used to train the five-fold dataset, and early stopping was implemented to control overfitting by monitoring the validation log loss. The test data was predicted using the five-fold model, and test-time augmentation was applied to ensure confident predictions. The model's performance was tracked using WANDB.