| 摘要: | 自 2020 年以來，整個社會掀起一波房地產討論熱潮，房屋、土地買賣移轉等相關話題此起彼落。在房地產關注度如此高之際，相關產業亦需加速數位轉型的腳步，透過持續優化內部作業流程、以演算法及大數據，建構客戶關係管理系統等，以新思維帶領企業邁向瞬變萬千的挑戰，「房地產科技」成為新的發展重點。其中，待銷售的房屋，對於房地產業者來說即為公司主要之獲利來源，然而，目前受委託的物件並無一個客觀用以衡量其競爭力的依據，若能開發相關工具或指標來衡量物件狀態，將有助於提升物件之銷售效率。經由過往文獻回顧發現較無客觀評斷房屋競爭力，關於銷售天數的研究與探討。因此，本研究之目的如下：
1. 運用不動產仲介業承接之房屋成交資料來建構銷售天數模型，並評估準確率。
2. 找出能有效預測房屋銷售天數之顯著影響因子與變數。
本研究基於房地產產業的應用背景，使用房地產委託之原始物件資料整理後，以演算法篩選重要特徵、運用各種方法進行建模並比較優劣，並以十摺交叉法作為驗證評估各種模型效益。經模型效益評估後，隨機森林及梯度提昇機的表現較線性回歸為佳，在評估回歸指標MSE與MAE 的結果上亦同。在重要的影響變數探討上，在自變數與銷售天數預測模型之間的關係，以「大門朝向」、「行政區」、「賣屋原因」為最主要影響因素的三大變數。;Since 2020, there has been a wave of real estate discussion in the society, and there are many topics related to housing and land sales and transfers. With such a high level of interest in real estate, the industry needs to accelerate the pace of digital transformation by continuously optimizing internal workflows, building customer relationship management systems with algorithms and big data, etc., and leading companies to the ever-changing challenges with new thinking. However, there is currently no objective basis for measuring the competitiveness of the objects commissioned. If relevant tools or indicators can be developed to measure the status of the objects, it will help to improve the efficiency of the sales of the objects. A review of past literature reveals that there is no objective assessment of housing competitiveness, and studies and research on days of sales. Therefore, the objectives of this study are as follows.
1. to construct a days-to-sale model and evaluate the accuracy rate by using the data of housing transactions undertaken by the real estate brokerage industry.
2. to identify the significant factors and variables that can effectively predict the days of sales.
Based on the application background of real estate industry, this study uses the original object data commissioned by the real estate industry and then uses an algorithm to filter the important features, apply various methods to model and compare the advantages and disadvantages, and use the cross-fold method as a validation to evaluate the effectiveness of various models. After the evaluation of model effectiveness, the random forest and gradient lifter performed better than linear regression, and the same results were obtained for the regression indicators MSE and MAE. In terms of important variables, the relationship between the independent variables and the day-of-sale prediction model, the three most important variables were "gate orientation", "administrative area", and "reason for sale". |
