Crowdfunding is a methodology of funding a project from a large number of people. With the Internet and social networking, this type of funding rapidly gained popularity. However more than 60% of projects are not funded, thus it is necessary to prepare carefully the crowdfunding campaign. Moreover, during the campaign, it is critical to estimate the success as soon as possible in order to react adequately (reorganization, communication): prediction tools are then essential. In this article, we propose several methods to improve the prediction of the amount raised during a crowdfunding campaign using the k-NN algorithm. The first proposition consists in using a clustering algorithm in order to segment the learning set and to facilitate the scaling for big data sets. The second proposition consists in extracting relevant features from the time series and information on the campaigns, in order to have a vector representation.