Multimodal ML · prediction
Amazon ML Challenge - Product Price Prediction
An ensemble product price prediction model combining CNNs, DNNs, and natural language processing. Sentence Transformers, XGBoost, and multimodal feature fusion were used to predict prices from product descriptions and images, achieving an SMAPE score of 56 and a top-percentile ranking.
Outcome
Achieved an SMAPE score of 56.
Evaluation
The model was evaluated with the challenge SMAPE metric and reached a top-percentile ranking.
Context
Product price prediction requires combining information from descriptions and images rather than relying on one modality in isolation.
Contribution
Combined CNN, DNN, and NLP signals with Sentence Transformers and XGBoost in an ensemble pipeline.
Architecture and workflow
Image features · Product text embeddings · CNN and DNN models · Sentence Transformers · XGBoost ensemble
Limitation
Challenge performance should not be read as a universal production pricing guarantee; the public record is limited to the competition setting.