| dc.contributor.advisor | 楊建民<br>洪為璽 | zh_TW |
| dc.contributor.advisor | Yang, Jiann-Min<br>Hung,Wei-Hsi | en_US |
| dc.contributor.author (作者) | 謝佳穎 | zh_TW |
| dc.contributor.author (作者) | Hsieh, Chia-Ying | en_US |
| dc.creator (作者) | 謝佳穎 | zh_TW |
| dc.creator (作者) | Hsieh, Chia-Ying | en_US |
| dc.date (日期) | 2018 | en_US |
| dc.date.accessioned | 30-七月-2018 14:55:08 (UTC+8) | - |
| dc.date.available | 30-七月-2018 14:55:08 (UTC+8) | - |
| dc.date.issued (上傳時間) | 30-七月-2018 14:55:08 (UTC+8) | - |
| dc.identifier (其他 識別碼) | G0104356021 | en_US |
| dc.identifier.uri (URI) | http://nccur.lib.nccu.edu.tw/handle/140.119/119035 | - |
| dc.description (描述) | 碩士 | zh_TW |
| dc.description (描述) | 國立政治大學 | zh_TW |
| dc.description (描述) | 資訊管理學系 | zh_TW |
| dc.description (描述) | 104356021 | zh_TW |
| dc.description.abstract (摘要) | 財政健全乃國家經濟永續發展之基石，政府為能合理、有效地利用資源，每年會依據可能之歲出及歲入情形來編列預算，以為來年之資金需求預作準備；而賦稅收入為我國政府最主要的歲入來源，因此，如何有效地估測賦稅收入以編列歲入預算，是政府重要的課題。以往已有利用各種如迴歸分析、時間序列分析方法預測稅收的研究，然而其中大部分的研究都係針對政府整體的年收入，或是以各項國稅為主，較少以地方政府及各月收入為標的進行預測；此外，隨著資訊科技的發展，各種硬體的運算效能一日千里，讓類神經網路又再次受到重視，且使以往受限於硬體效能的類神經網路架構得以實現。在近幾年興起的類神經網路中，遞迴式類神經網路運用記憶單元來傳遞資訊，在時間序列相關運用上取得重大的進展，而其改良版本─長短期記憶遞迴式類神經網路是最主要被運用的架構。本研究以臺北市、新北市等六個縣市的土地增值稅之增減及高低作為預測目標，系統性地組合不同的神經網路超參數，逐步找出最適合的神經網路架構，並分別針對各縣市之稅收增減及稅收高低進行預測；經實證研究，本研究所採用之長短期記憶遞迴式類神經網路架構在稅收增減及稅收高低的預測上，平均及最佳準確率分別達65%及69%，單一縣市最佳的準確率則分別為76%及93%。另外，在資料分布對神經網路的影響方面，經探討各縣市的預測結果，發現在稅收增減方面，資料的分布對於模型預測無明顯影響，而稅收高低的資料分布若差異越大，則預測的結果越準確。 | zh_TW |
| dc.description.abstract (摘要) | Financial integrity is the cornerstone of the country`s economic sustainable development. In order to use the funds reasonably and effectively, the government will compose budget according to the possible annual revenues and annual expenditures, so as to prepare for the funding needs of the coming year.Tax revenue is the most important source of revenue for our government. Therefore, it`s an important issue for the government to estimate the tax revenue effectively.There are various studies such as regression analysis and time series analysis have been used to predict tax revenue in the past. However, most of the research is based on the overall government`s annual income or national taxes, and lack of local government and monthly income. In addition, with the development of information technology, the computing power of various hardware has been steadily increasing. The neural network has received attention once again, and the architecture that has been limited by hardware performance also can be realized.In the neural network that emerged in recent years, the recurrent neural network uses memory cells to transmit information, making significant progress in time-series related applications. And the improved version - long short-term memory recurrent neural networks is the main architecture be used. In his research, we uses the increase and decrease of land value increment tax in six counties such as Taipei City and New Taipei City as the forecast target. We test different neural network hyperparameters systematically to find the most suitable neural network architecture gradually, and forecast the tax increase and decrease and the tax level of the six cities respectively.According to empirical research, the long short-term memory recurrent neural network architecture adopted in this research has an average and best accuracy of 65% and 69% in terms of tax increase and decrease and tax level, respectively.The best result of accuracy rates for a single county are 76% and 93% respectively.Otherwise, in terms of the impact of data distribution on neural networks, after exploring the forecast results of various cities. We found that the distribution of data has no significant impact when predict tax increase and decrease. But the accuracy rate will increase when the distribution of tax revenue level is less balanced. | en_US |
| dc.description.tableofcontents | 第一章 緒論 1第一節 研究背景與動機 1第二節 研究目的 4第二章 文獻探討 5第一節 時間序列分析之稅收預測研究文獻 5第二節 遞迴式類神經網路 8第三節 LSTM應用於時間序列預測相關文獻 11第四節 小結 13第三章 研究方法 14第一節 研究流程 14第二節 實證資料內容與資料預處理 15第三節 LSTM類神經網路模型設計 19第四節 實證資料處理 22第五節 模型評估準則 23第四章 實證結果及分析 24第一節 神經網路超參數調整 24第二節 六都稅收預測 31第三節 稅收預測結果分析 40第五章 結論與建議 42第一節 研究結論 42第二節 研究限制 44第三節 研究建議 45參考文獻 46英文部分 46中文部分 47 | zh_TW |
| dc.format.extent | 1494789 bytes | - |
| dc.format.mimetype | application/pdf | - |
| dc.source.uri (資料來源) | http://thesis.lib.nccu.edu.tw/record/#G0104356021 | en_US |
| dc.subject (關鍵詞) | 土地增值稅 | zh_TW |
| dc.subject (關鍵詞) | 長短期記憶遞迴類神經網路 | zh_TW |
| dc.subject (關鍵詞) | 稅收預測 | zh_TW |
| dc.subject (關鍵詞) | Land value increment tax | en_US |
| dc.subject (關鍵詞) | Long short-term memory recurrent neural network | en_US |
| dc.subject (關鍵詞) | Tax forecasting | en_US |
| dc.title (題名) | 利用長短期遞迴類神經網路建構地方政府稅收預測模式之研究─以土地增值稅為例 | zh_TW |
| dc.title (題名) | Study on the Constructing Forecasting Model of Local Government Revenue using LSTM-RNN Taking Land Value Increment Tax as an Example | en_US |
| dc.type (資料類型) | thesis | en_US |
| dc.relation.reference (參考文獻) | 英文部分1. Felix A. Gers ,Jurgen Schmidhuber ,Fred Cummins. (1999, January). Learning to Forget: Continual Prediction with LSTM. Technical Report IDSIA-01-99.2. Gao, Q. (2016, 5). STOCK MARKET FORECASTING USING RECURRENT NEURAL NETWORK.3. Hansson, M. (2017). On stock return prediction with LSTM networks. Lund: Department of Economics Lund University.4. Hochreiter S. ,and Schmidhuber J. (1997 9(8)). Long short-term memory. Neural computation, pp. 1735-1780.5. L.Bobbou and O. Bousquet. (2007). The tradeoffs of large scale learning. In Proceedings of 20th International Conference on Neural Information Processing Systems(NIPS), pp. 161-168.6. Luca Di Persio , Oleksandr Honchar. (2017). Recurrent neural networks approach to the financial forecast of Google assets. INTERNATIONAL JOURNAL OF MATHEMATICS AND COMPUTERS IN SIMULATION Volume 11, 2017, pp. 7-13.7. Luca Di Persio,Oleksandr Honchar. (2016, January). Artificial neural networks approach to the forecast of stock market price movements. International Journal of Economics and Management Systems.8. Nijolė Maknickienė, Aleksandras Vytautas Rutkauskas, Algirdas Maknickas. (2011). Investigation of financial market prediction by recurrent neural network. Innovative Infotechnologies for Science, Business and Education 2(11), pp. 3-8.9. Ronald J. Williams,David Zipser. (1995). Gradient-based learning algorithms for recurrent networks and their computational complexity. In Back-Propagation: Theory,Architectures and Applications (pp. 433-486).10. Sepp Hochreiter. (1998). Recurrent Neural Net Learning and Vanishing Gradient. International Journal of Uncertainty, Fuzziness and Knowledge-Based Systems 6(2), p. 107{116.11. Sreelekshmy Selvin, Vinayakumar R, Gopalakrishnan E.A, Vijay Krishna Menon, Soman K.P. (2017). Stock price prediction using LSTM, RNN and CNN-sliding window model. International Conference on Advances in Computing, Communications and Informatics (ICACCI).中文部分1. 江枝華. (2003). 所得稅稅收預測及其管理之研究. 台北市: 國立政治大學社會科學學院.2. 李川源. (1979). 台灣地區所得稅稅收之預測. 新竹市: 國立交通大學管理研究所.3. 陳昶憲、吳清俊、鍾侑達. (2004). 遞迴式類神經模式於日流量預測之應用. 中華水土保持學報, 頁 187-195.4. 彭琇嫦. (2008). 賦稅收入預測模型之研究. 台北市: 天主教輔仁大學應用統計研究所 碩士論文.5. 黃美玲. (2011). 建構所得稅預測模式之研究. 台中市: 國立臺中技術學院企業管理系事業經營碩士班.6. 蔡招榮. (1979). 台灣財產稅月稅收預測隨機時間序列之應用. 新竹市: 國立交通大學管理研究所.7. 鄭允中. (2017). 基於長短期記憶遞迴神經網路之新台幣兌美元匯率預測模型. 台北市: 國立臺灣大學電機資訊學院資訊工程學系 碩士論文. | zh_TW |
| dc.identifier.doi (DOI) | 10.6814/THE.NCCU.MIS.010.2018.A05 | - |
