In this project, our data source is the stakeholders who are Jouyu’s family business. It is kind of transport industry. The business model is to transport customers’ excavators to the destination they require through the trailers whom the stakeholders own. However, since the stakeholders are used to manually record the trips that customers require, they predict the next demand only based on their experience. Without the assistant of technology, stakeholders are unable to anticipate the demand of trailers in the future accurately.
Based on these business problems that the stakeholders meet, we set two forecasting goals. One is to forecast the trend of small and large trailers. We will use the data of monthly trips to estimate the demand of different types of trailers. The other one is to forecast the trend of monthly demand for two types of customers. The reason is that we attempt to find out the biggest potential demand in this industry so that we can offer some customized service. In addition, our goal is prospective, because we add data month by month. As for time series, we have four time series, including customers (company, and individual) and excavators (small, and large). Overall, our purpose is to forecast the demand in 2015.
When it comes to the data, we totally receive 3682 data from the stakeholders. The interval of data is 4 years, from 2011 to 2014. The data columns include date, customer types (company, and individual), and trailer types (small, and large). In order to know the situation of data, we visualize the data through tableau. The data are divided into 2 groups. One plot is for customers (company vs individual), and the other is for excavators (small vs large). It is found that there is a trend but no seasonality. Additionally, after partitioning the data, we have 3 years for training period (2011/1~2013/12) and 1 year for validation (2014/1~2014/12).
After partition, we try to analyze the data through naïve, and holt-winters. Nevertheless, the results perform not well. Therefore, we opt for moving average, exponential smoothing and neural network to rerun our data. Simultaneously, we use neural network to get forecasting value of year 2015. The error rate is regarded as benchmark. It is found that the error rate of neural network is smaller than that of moving average and exponential smoothing. Hence, we decide to choose neural network to predict the demand in 2015, and the charts of results are attached on item 6 and 7.
In conclusion, according to the prediction that we get, we only can recommend Jouyu’s family that they may negotiate with the customers about the short-term contract. From the stakeholders’ perspective, they can keep the valued customers. For the customers, they can get some discount. Through this way, stakeholder can build the strong relationship with customers.