近年來討論家外媒體（Out-of-Home, OOH）具單向溝通的限制，容易讓消費者視而不見、廣告主盲目的猜測目標受眾（Target Audience, TA）及缺少直接連結與購買商品的介面，這導致了廣告主行銷效率低落甚至流失既有客戶。現今數位行銷的致勝關鍵，是如何更有效地規劃後續行銷活動以及提升行銷效率，而不再是單純地增加新客戶或解析僅能呈現銷售數據的報表。消費者的喜好與市場千變萬化、與消費者溝通的廣告媒體已日漸飽和、以及無用的廣告內容造成消費者越來越反感，綜上所述，如何改善消費者體驗將會是數位媒體行銷所面臨的一大課題。
因此本研究蒐集歷史脈絡統計資料，依據生活型態理論來分析消費者特徵，包括對節目喜好的行為特徵，使用決策樹的特徵提取逐步提高模型性能；並提出一個隨機森林預測模型，透過滾動迭代步驟調整模型來預測消費者商品喜好，藉以調整媒體所投放節目與商品的關聯性或行銷策略，達到精準行銷的效果。
本研究結果顯示在數據較少的情況下，仍可有效地將商品推薦給可能對它感興趣的消費者。零售業者透過此資料科學的方法，不僅可運用人工智慧（Artificial Intelligence, AI）分析大量且複雜的數據以更全面地觀察TA輪廓，更可快速地預測消費者喜好；而作為OOH的廣告主，使用本模型預測結果以調整OOH平台節目置入性行銷及廣告投放的參數組態，更可精準地客製化行銷商品。隨著日後模型逐步完整且準確，與消費者良好的互動及商品的強聯結，以此打破傳統OOH的限制，創造出即時行銷與動態貼近消費者需求的最佳化廣告解決方案。

In recent years, the limitations of Out-of-Home advertising (OOH) with one-way communication has been discussed, including easily causing consumers to turn a blind eye, blindly guessing the Target Audience (TA), and lack of interface for browsing and buying goods. This has caused the advertisers to be inefficient in marketing and even customer churn. The key to success in digital marketing today is how to plan follow-up marketing activities more effectively and improve marketing efficiency, rather than simply adding new customers or analyzing reports that only present sales data. The ever-changing market and consumer preferences, to communicate with consumers has become increasingly saturated media advertising and unwanted advertising content consumers are increasingly caused resentment, summary, how to improve the consumer experience of digital media marketing will be faced One of the major issues.
Therefore, this research collects historical statistical data, analyzes consumer characteristics based on Lifestyle theory, including behavioral characteristics of program preferences, and uses the feature extraction of decision trees to gradually improve model performance; and proposes a Random Forest prediction model through rolling iterations Steps to adjust the model to predict consumer product preferences, so as to adjust the relevance of the program and the product placed by the media or the marketing strategy, to achieve the effect of precise marketing.
The results of this study show data in the case of small, goods can still be effectively recommended to the user may be interested in it. Through this Data Science method, retailers can not only use Artificial Intelligence (AI) to analyze a large amount of complex data to observe the TA profile more comprehensively, but also to quickly predict consumer preferences; As an OOH advertiser, use the prediction results of this model to adjust the parameter configuration of OOH platform program placement marketing and advertising, and more accurately customize marketing products. As the future model is gradually complete and accurate, good interaction with consumers and strengthened connection with products will break the limitations of traditional OOH, and create real-time marketing and an optimized advertising solution that dynamically meets consumer needs.

[1]Abellán, J., & Castellano, J. G., (2017), A comparative study on base classifiers in
ensemble methods for credit scoring, Expert Systems with Applications, p.1-10
[2]Leo Breiman, (2001), Random Forest, Machine learning, p. 5-32
[3]Hochreiter S., & Schmidhuber J, (1997), Long short-term memory, Neural computation, p.1735-1780
[4]LeCun Y., Bengio Y., & Hinton G, (2015), Deep learning, nature, 521(7553), 436-444
[5]Rasmus Houlind & Colin Shearer, (2019), Make It All About Me: Leveraging Omnichannel and AI for Marketing Success, Lid Pub Inc
[6]Kelly G. A., (1955), The Psychology of Personal Constructs, New York
[7]Klein G. S., (1970), Perception, motives, and personality, Alfred A, Knopf
[8]Koehn, D., Lessmann, S., & Schaal, M, (2020), Predicting online shopping behaviour
from clickstream data using deep learning, Expert Systems with Applications,p. 150, 113342
[9]Philip Kotler, Hermawan Kartajaya, & Iwan Setiawan, (2016), Marketing 4.0: moving
from traditional to digital, Wiley
[10]Gerry Murray, Laurie Buczek, Wayne Kurtzman, Amy Loomis, Ph.D., & Alan
[11]Webber, (2020, 10), IDC FutureScape: Worldwide B2B Sales Leaders 2021 Predictions
[12]Pal, M, (2005), Random forest classifier for remote sensing classification,
International Journal of Remote Sensing, 217-222
[13]Plummer, J. T, (1974), The Concept and Application of Life Style Segmentation,
Research Article
[14]Raschka, Sebastian, & Vahid Mirjalili, (2017), Python Machine Learning. Packt
Publishing
[15]Reynolds, Fred D., & William R. Darden, (2011), Construing life style and psychographics, Marketing Classics Press
[16]Verhoef, Peter C., Pallassana K. Kannan, & J. Jeffrey Inman, (2015), From Multi-
Channel Retailing to Omni-Channel Retailing: Introduction to the Special
Issue on Multi-Channel Retailing, Journal of Retailing, Pages 174-181
[17]William D. Wells, (1974), Lifestyle and Psychographs Paperback, Amer Marketing
Assn
[18]Zhai X., Shi P., Xu L., Wang Y., & Chen X, (2019), Prediction Model of User
Purchase Behavior Based on Deep Forest, Computer Science, p. 190-194
[19]Zocca, Spacagna, Slater, & Roelants, (2017), Python Deep Learning, Packt 39
[20]未來流通研究所, (2020 年12 月08 日), 台灣實體零售業的網路銷售大追擊, 擷取自 未來流通研究所: https://www.mirai.com.tw/taiwan-phyisical-retailindustry-transformation-analysis-20201208/
[21]江孟錡, (2020), 零售企業發展虛實融合之探索性研究, 國立臺灣大學
[22]行政院, (2020), 台灣經濟發展的利多與機會報告
[23]黃心瑜, (97), 台北市長青學苑高齡者觀看電視戲劇節目與生活型態關係之研究,國立臺灣師範大學