Demand Forecasting and Fast Mover Prediction

Nagadevara, V.P, Wadhwani N, Khinvasara U, Kothari M, Agarwal M, Arora H.S

Today, at USD 1.7 trillion, the global apparel market is one of the fastest growing ones expected to be around USD 2 trillion by 2028.² In fashion industry, forecasting and inventory management both tend to be complex tasks due to short product life cycles and high volatility of demand. Sales in the fashion industry are difficult to predict because they are influenced by weather, seasonal demands, trends, marketing strategies and other cultural factors.

A project was undertaken by students for a leading fashion retailer that aimed to extract relevant product attributes from photos, forecast units sold for each product object, and identify/ classify the fast-moving products by leveraging computer vision and machine learning algorithms. By adopting this data-driven strategy for feature extraction, unit sales forecasting and fast mover identification, team tried to predict to maximize sales and avoid stockouts.

The team divided the project into 3 parts: Product Image Classification for feature extraction, Regression Analysis to estimate sales quantity and, Fast Mover classification for inventory data.

In part 1, the team created a neural network model built upon MobileNetV3 architecture integrating a self-attention layer and single shot detection. The model was trained to extract features like fit, waist rise, collar etc. from product images. The performance of the model was evaluated based on its accuracy in identifying product attributes.

In the second part, Random Forest Regressor was used to focus on demand forecasting at both India and store level. The accuracy of different forecasting models was evaluated using Absolute Percentage Error (MAPE) in predicting sales quantities. This enabled informed decision-making in production planning, inventory management and sales optimization.

In the concluding part, the team introduced inventory data to utilize a Gradient Boosting Classifier to identify fast-moving items. In order to minimize the risk of overlooking high-moving items and to ensure they are accurately identified, ‘Recall’ was used a metric enabling effective inventory management and resource allocation.

The team provided a comprehensive analysis of product features, sales forecasting and fast mover identification to the company. They leveraged advanced techniques and metrics to derive valuable insights that support decision- making related to inventory management and sales optimization.

The recommendations made by the team can substantially contribute to improving the efficiency and profitability to the company’s India operations in the very dynamic and highly competitive fashion industry.

References:

1- Nagadevara, V.P., Wadhwani N., Khinvasara U., Kothari M., Agarwal M., Arora H.S

2- https://www.statista.com/topics/5091/apparel-market-worldwide/#topicOverview