The data set is available on Kaggle at https://www.kaggle.com/ravirajsinh45/real-life-industrial-dataset-of-casting-product
Around 7000 augmented images were provided, which were divided into defective and non-defective images.
Train and test datasets were already divided in a 0.90
percentage. Additionally 1000 more un augmented images were provided.
I have augmented them using the general augmentation techniques and started the project using statistical ML models.
SVM RBF and Random forest gave the best results out of all, with SVM giving 99%.
Next phase is using Neural networks to achieve near 100% accuracy.