Price Elasticity Prediction by Cloud Machine learning technique – the Case of Tiny E-commerce Company in Taiwan

• Main Authors: To Shia, 夏鐸
• Other Authors: Min Hsin Huang
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/m52w4r

碩士 === 國立中山大學 === 企業管理學系研究所 === 106 === This article is about to prove that a tiny e-commerce company can analyze online trading data and give business management valuable insight. To compete with giant e-commerce company, tiny company needs to fit fast-changing modern business world by flexible strategy and functional differentiation. In aspect to limited resource, general tiny company didn’t have enough resource to fight with their strong competitors. Since obviously progress in machine learning (ML) field recently, entry barriers of ML is not as high as before. We wonder that if tiny company can gain good quality predictive results by analyzing key managerial numbers or trading data in ML way as their richer competitors. Thus, we design a tour of experience completely step-by-step ML procedures with real trading datasets which provided by a small hot copy (printer) consumables supplier in Taiwan. In order to simulate lack of resource situation, we won’t use high-end computers, write programming codes or demonstrate complex algorithms during whole process. In the middle of process, we will run Microsoft Azure machine learning studio cloud computing technology to help us simplify a part of ML flow and cut the time of calculating. In this case, we choose price elasticity as predict target, analyzing the relationship with brand, price, platform and seasonal factors. Follow steps of Azure ML, from data receiving, preprocess, feature define, apply algorithm, evaluate model to develop model. In process, we also compared several algorithms simultaneously to evaluated best target prediction quality. After obtaining prediction result, we use Microsoft POWER BI to visualize result and establish inter-reactive dashboard for both desktop and mobile device. As a result, we proved even tiny company can use cloud ML technology to build predictive model by itself without limits. By cloud computing ML method, it cuts ML cost into acceptable level for tiny company or individuals. As data collecting and Algorithm continued steadily grow into well-developed technology, next dilemma of ML will move to result explanation. In a future, industrial and humanities-related expert will play an important role in ML field.