Handwritten Signature Verification and the S&P 500 Index Forecasting by Artificial Neural Network

• Main Authors: Kao, Chienchih, 高健智
• Other Authors: Chuang, Shangjen
• Format: Others
• Language: zh-TW
• Published: 2012
• Online Access: http://ndltd.ncl.edu.tw/handle/32922463160735521984

碩士 === 國立高雄海洋科技大學 === 電訊工程研究所 === 100 === Handwritten signature verification system use personal signature as a recognizing tool, which records signature and analyzes person’s handwriting, according to different personal handwritten way. To form the complete handwritten signature verification system, Probabilistic Neural Network, image preprocess, and feature analysis technology were used in this research. The database of this system was consisted of 20 people’s personal handwriting; use image process technology to decrease the variability which might influence the classification, and analyze the extract feature by using statistical feature. This feature analysis way can make network classifier feature and training much easier, and also can increase handwritten signature recognition rate. S&P 500 Index System use Back-propagation Neural Network and Echo State Networks as the tools to build up models. The goal is to predict the highest-price, lowest- price, and close price in the daily S&P 500 Index, and also make a comparison between Back-propagation Neural Network and Echo State Networks. The data is collected between 4th Jan, 1965 to 22nd Oct, 2007, 10767 days in total. The chosen data include daily open-price, a day before, and seven days prior’s open-price, highest-price, lowest-price, the day of the week, and trading volume as neural network input variable. The result shows that the stock prices increase rapidly from 1995 to 2007. This caused the result that the network cannot predict a better-forecast value. Therefore, this paper categorize S&P 500 into two models, one is training between 1965-1990 and testing between 1995-2000, while another one is training from 1995-2000 and testing from 2001-2007.