Abstract:
|
One of the most prominent application fields is Computer Vision and Image Processing that can quickly extract large amounts of "crude" data or information from images such as surveys and medical images. Furthermore, machine learning algorithms can be applied to extract and transform data into a structured format for data entry validation. Subsequently, it reduces the costs for conducting such validation. This application involves several challenges: 1) difficulty to extract user input and filter out noise; 2) various layouts that can complicate the retrieving process. The layouts could be from the simplest (a single layout for each question) to the most complex (a mixture of Likert-point scale, open-end, and skip-patterned questions). In this project, we sought to use Support Vector Machine (SVM) to extract information from scanned hard-copies of the MFI-20, SF-36, PROMIS instruments. Our preliminary results have successfully demonstrated 90% agreement with the MFI-20 data entered in MS ACCESS Form. Thus, only 10% requires for validating manually. As a result, SVM could be considered a cost-effective tool for data entry validation.
|