Hi @Mandru Venkata Durga Prasad,
Thank you for reaching out to Microsoft Q&A forum!
It seems that your custom-trained image moderation model is incorrectly flagging non-incident images as incidents. To address this, review the diversity and accuracy of your training data, consider adjusting the model's threshold, and retrain the model with these improvements. Additionally, you might want to test the model with a broader range of images and consider using the Standard API instead of the Rapid API, as it might offer more reliable results.
I hope this helps! Thank you.