SecondhandAI (Price Forecast for Secondhand Marketplaces)

Welcome

This team's thematic problem is second-hand market item pricing. We have created a product that aims to provide buyers and sellers with competitive pricing information and improve shopping experiences. The online resale market has grown significantly in the past two years, culminating the issues related to the pandemic, inflation, and increased environmental awareness. A price forecast in a secondhand marketplace benefit both sellers and buyers by promoting transparency, competition, efficiency, and economic growth.

Our Mission

To empower second-hand market customers with data-driven price recommendations for electronics, fostering transparency and enriched experiences​​

Key Learnings and Impact

The online resale market has been booming in recent years. According to CNBC, the resale and secondhand market is expected to reach $53 billion by 20231. This has been fueled by shifting consumer demands, from shopping in more sustainable ways as well as trying to secure hard-to-find luxury items. The rise of the secondhand marketplace has been driven by online resale platforms such as Ebay. A price forecast in a secondhand marketplace benefits both sellers and buyers by promoting transparency, competition, efficiency, and economic growth.

Examples

Fair Value "USED" Predicted Price: $115

Current market price: link

How It Works

SecondhandAI was developed using AWS tools. We were able to utilize various components (including Cloudwatch, AWS Lambda, S3, and Sagemaker) to develop our end-to-end pipeline. Our models were primarily developed on Sagemaker instances in order to scale out and scale up whenever necessary to run large language and random forest classification models.

Our application is powered by a random forest classification model with various machine learning techniques built in for feature engineering. The model is pre-trained on an dataset gathered from Ebay, focusing on all electronics data from various categories including Computers/Tablets, Cameras/Photos, Consumer Electronics, Cellphones & Accessories, and Video Games. Natural Language Processing techniques and various unsupervised and supervised techniques were used for feature engineering to ensure accurate price prediction.

Evaluation

When evaluating the model performance, we are predicting the price of a continuous variable. We utilized a combination of three metrics - RMSE, MAE, and WMAPE. These metrics measure accuracy by comparing predicted prices to actual prices. RMSE is useful for larger errors, while MAE is less sensitive to outliers. WMAPE is ideal for a percentage-based focus with greater emphasis on larger errors. Lower values for all three metrics indicate better performance. Employing this trio provides a more comprehensive evaluation of the model's performance.

Acknowledgements

We would like to acknowledge our instructors, Joyce Shen and Zona Kostic, for supporting us in the development of our project and application as well as the rest of our Spring 2023 W210: Section 1 cohort! We would also like to thank the students from W231 who participated in an Ethics and Privacy review of our application. We gained valuable insights from our peers and instructors that we have implemented within our application.