研發適用於公司營收預測系統的智慧型數學建模方法

Abstract

本產學合作研發計畫，由交通大學智慧型計算實驗室與嘉實資訊股份有限公司合作，以嘉實資 訊擁有巨量的商業資訊源和主持人擁有的自動化特徵選取與建模技術，設計研發適用於公司營收預測 系統的智慧型數學建模方法，期能解決應用機器學習方法於預測營收時面臨的雙目標組合最佳化問 題，即使用最少量的穩定重要特徵即獲得最小的預測誤差。準確預測公司營收能進一步發展預測每股 盈餘或股價等資訊，更有效反應股價的變化。本計畫擬設計同時進行特徵選取與建模的自動化方法， 以主持人研發之最佳化核心技術：1)直交實驗設計與因子分析以及2)繼承式雙目標基因演算法為基 礎，來評估模糊系統、類神經網路與適應性網路架構的模糊推論系統、支援向量回歸模型等方法設計 能同時兼顧預測正確性與自動特徵選取的最佳化高效能建模方法。本研究主題理論與實務並重，預計 先以統一超商與全家超商為公司範例，主要的研究目標簡述如下： 1. 研發適用於公司營收預測系統的智慧型數學建模方法。 2. 應用此方法於建置統一超商與全家超商之營收預測系統之核心元件。 3. 為此方法的核心元件建立與嘉實資訊之資訊服務系統互動之介面。 4. 提供介面使嘉實資訊之資訊服務系統能提供核心元件輸出的營收預測與其所選取之重要財經特 徵的相關資訊。

The project, “Developing an intelligent mathematic modeling method for designing a prediction system of companies’ revenue”, is a collaborated by the principal investigator of Intelligent Computing Lab. Of National Chiao-Tung University and SysJust Crop.. This project intends to solve the bi-objective combinationary optimization problem on the revenue prediction of companies by designing an automatic feature selection and mathematic modeling method, by utilizing the large business information owned by SysJust and the automatic feature selection and modeling techniques provided by the principle investagor, which is to selected the minimal stable and important feature set to acquire the minimized prediction errors. To accurate prediction of revenue can be further developing to predidct the earning per share and the share prices, and response to share prices rapidly. This project proposes a method to simultaneously minimize the number of selected feature and model tuning. This method will be integrated with two of our developed core techniques, which are 1) orthogonal experimental design and factor analysis and 2) inheritable bi-objective genetic algorithm. Furthermore, this efficient method will applied to assess Fuzy system, Neural Networks and Adaptive Neuro-Fuzzy Inference System, Support Vector Regression models for designing this optimization and modeling method. This project uses companies like Seven-EleveN and Family Mart, as examples, the main goals of this project are summarized as follows: 1. Developing an intelligent mathematic modeling method for designing a prediction system of companies’ revenue. 2. To apply this method to construct the core componiments of revenue prediction systems for Seven-EleveN and Family Mart. 3. To provide the interface for interacting with SysJust’s information system for these core componiments. 4. To provide the interface for output the revenue prediction and the selected important financial information within SysJust’s information system for these core componiments.