| Summary: | 碩士 === 國立交通大學 === 工業工程與管理系所 === 103 === Recently, new kinds of smart devices are invented and sold to the market rapidly, and we witness a rise of wearable devices. Since the release of wearable devices by companies, we found that the sales of most products are below expectation. We thought that the fundamental reason is that wearable device are still in the introduction stage of its product life cycle, and the market demand is still ambiguous. In order to find the demand of wearable device, this study uses data mining techniques to find the potential consumers’ background and the key attributes which cause them to purchase. And this study also proposed an approach to achieve product recommendation.
This study conducts correspondence analysis to find the relation of function attribute and different kind of wearable device. We found that smart wristbands should be provided with sleep pattern monitor, heart rate monitor, burnt calories tracker, body temperature sensor and pedometer. Smart watches should be provided with audio recording, video recording, GPS receiver, phone function, message reminding, music player, touch screen and voice control. Sport watches should be provided with heart rate monitor, burnt calories tracker, pedometer, odometer, speedometer and barometer.
We also conduct multiple conducts correspondence analysis in order to investigate the relation between consumer and different kinds of wearable devices. The following is the result of multiple conducts correspondence analysis. For smart wristbands, we suggest the manufactures should place their target market at consumers who are above 40 years old, female or exercise frequency is low. For smart watches, the target market should be placed at consumers who are under 19 years old, male or exercise frequency is normal. As for sport watch, the target market should be placed at consumers who are 20~29 years old, male or exercise frequency is high.
To sum up, in variable filtering part, this study compares association rules and correspondence analysis. And in classification part, this study uses k-nearest neighbors method, with 66% of prediction accuracy, which is capable of product recommendation.
|
|---|
