As part of my super specialization in analytics at IFIM, I got introduced to deep learning models and an opportunity to work on this project.
About the Project
As part of this project, we must build an image classification system using Convolutional neural networks for classifying the images into 10 categories of fashion garments/accessories, followed by a recommendation system. The recommendations are based on the type of garment/accessory, price, and previous ratings.
When an image and price of a fashion garment/accessory is given as an input, the model must first analyze and classify the image and then provide suitable recommendations.
We are using a sample to 30000 images from the Fashion MNIST dataset from Keras.
Building the Image recognition model.
We built the classification model in Google Colaboratory using the Sequential API from Keras.
The image classification model achieved 91% accuracy in classifying the Fashion garments/accessories.
When an image of a fashion garment/accessory and its price is given as input, the image is classified by the CNN model into the relevant product category.
The recommendation system was able to give out appropriate recommendations based on the product category, price and previous ratings.
Takeaways and Future Scope
This model can be further improved by using image similarity along with the price of the product for giving recommendations instead of image classification. This model can be scaled up and be used by social media sites to recognize the design patterns in the images liked by the customers and suggest similar products that the customers can buy.
Mohammed Aamir Shuaib
Analytics and Marketing – Student at IFIM Business School