EXPLAIN THE FACTORS DECIDING THE CLASSIFICATION RESULTS IN AI MODELS FOR IMAGE ANALYSIS

Know More

CLIENT

One of the largest logistics and courier companies in the world

BUSINESS CHALLENGES

Increase customer satisfaction and reduce false damage shipment claims, by identifying damaged shipments at each transit point. The objectives were as follows:
• From a captured image of a parcel, predict if it is already damaged.
• Reduce false positives and thereby cut manual efforts and operational risk.

SOLUTION

A deep learning-based solution was created to identify the damaged parcels from the captured images.

The images were fed to a deep neural network-based model which extracted image features and then used them to classify the parcels as damaged or undamaged.
A local explainability framework was deployed to highlight the areas being considered by the model, to predict if the shipment was damaged.
The explanations were used to understand the kind of images that led to incorrect classification, thereafter debug the solution and increase its accuracy.

BUSINESS BENEFITS

Achieved an accuracy of more than 90% in classifying the parcels as damaged or undamaged.

The model was tuned to predict 100% damaged shipments correctly.