Over the past few years, crowdfunding platforms helped their users raise several billion dollars worldwide, thereby becoming a viable mechanism for people seeking funding to jump-start their business ventures. In spite of the widespread popularity and innovativeness in the concept of crowdfunding, many projects are still not able to succeed. Although, at the outset, the domain of crowdfunding appears to be intuitive and simple, it poses several new challenges from the analytics perspective due to the heterogeneous, complex and dynamic nature of the data associated with crowdfunding campaigns. In this project, we investigated the problem of predicting project success, which is multi-factorial and depends on a wide range of elements that are hard to characterize. To achieve this, we developed a systematic data-driven approach to resolve these challenges by utilizing vast amounts of historical data which can be leveraged to support crowdfunding project campaigns by predicting the success of projects. One of the important challenges in crowdfunding is to predict the success of the project using various kinds of project-related features. Existing approaches in data analytics for classification and regression cannot tackle this crowdfunding prediction problem since the goal here is to estimate the time for project success which is available only for a subset of projects (which succeeded in obtaining their target amount). Hence, the primary focus of this project was to incorporate the failed projects (which contain only partial information until the project end date) into the regression model, there is a need to develop new algorithms. The main goal of this project is to build accurate and robust prediction models for estimating project success in crowdfunding environments. We developed a unified probabilistic framework which integrates classification and regression. We also built an imputation model that calibrates the time to project success in an attempt to reduce the bias in the model estimators. This calibration step is performed using the estimated regularized inverse covariance matrix within an iterative convergence framework. In addition, we rigorously analyzed the important factors of crowdfunding to build novel algorithms that can overcome some critical drawbacks of existing approaches available in the literature. The project explored various kinds of complexities that arise in crowdfunding data and incorporated them into prediction models. A deeper understanding of the factors affecting investment decisions not only gave better success rate to the future projects but also provided better guidelines for project creators who will be interested in funding the projects.  We demonstrated that models which take into account both successful and failed projects during the training phase perform significantly better at predicting the success of future projects compared to the ones that only use the successful projects. We provided a rigorous evaluation using several sets of relevant features and show that adding few temporal features that are obtained at the project’s early stages can dramatically help in improving the overall performance. The main results of this work have been disseminated to the research community through publications, software, tutorials and other presentations. The main ideas and concepts developed in this project were also discussed in graduate-level computer science courses on data analytics. This project demonstrated the power of novel data analytics solutions in delivering better insights about various categories of real projects in crowdfunding environments by not only accurately estimating the chances of being successful but also quantitatively assessing the factors that are more responsible for bringing success. Though the proposed methods were primarily developed in the context of crowdfunding data, they can also be applied to various other forms of social data that is collected in other disciplines such as social science, engineering, and finance.

Last Modified: 08/01/2018
Modified by: Chandan K Reddy