Preliminary predictive and visual analytics for strategic business planning

Celero Solutions需要对战略业务规划的预测和可视化分析进行研究。一个特别令人担忧的问题是对加拿大经济，尤其是对加拿大信用合作社的业务做出预测。该研究项目的主要目标是：(1)设计和评估加拿大信用合作社业务的预测模型，该模型包含有关全球、国家和地方经济因素的公开可用和私人经济数据，以及(2)创建和评估一个仪表板，以易于理解的格式显示有关信用合作社健康和绩效的信息。各种各样的预测分析技术，包括统计回归、自回归综合移动平均线（ARIMA）、向量自回归（VAR）、机器学习（分类学习和深度学习）和数据挖掘，将被评估它们对预测任务的适用性。仪表板将被设计用于显示与经济模型相关的输入和汇总输出。用户将能够选择信用合作社业务的适当部分。将进行研究以确定图形用户界面（GUI）显示适当信息的适当方式，这取决于所考虑的细分市场。这些资料将提供给个别信用合作社，并在整个市场部分的综合价值范围内提出。在研究的评估部分，首先将使用一些人工数据对预测模型进行实施和验证。接下来，我们将把模型应用于历史数据，以确定其有效性。将使用训练-测试的方法，即只使用较旧的数据进行训练（创建模型），而使用较新的数据进行测试。另外，作为评估的一部分，我们将在仪表板上执行详细的软件测试，以确保所有所需的功能都存在，并且工作正确和可靠。

Celero Solutions requires research on predictive and visual analytics for strategic business planning. A particular concern is making predictions about the Canadian economy and especially the business of Canadian credit unions. The main objectives of the research project are (1) to design and evaluate a predictive model of the Canadian credit union business that incorporates publicly available and private economic data concerning global, national, and local economic factors, and (2) to create and evaluate a dashboard that displays information about credit union health and performance in an easy to understand format. A wide variety of predictive analytics techniques, including statistical regression, auto-regressive integrated moving averages (ARIMA), vector auto-regression (VAR), machine learning (classification learning and deep learning), and data mining will be evaluated for their suitability to the predictive task. The dashboard will be designed to display the input and summarized output related to the economic model. The user will be able to select an appropriate segment of the credit union business. Research will be performed to determine an appropriate way for the graphical user interface (GUI) to display appropriate information depending on the market segment under consideration. This information will be given for individual credit unions and also presented in the context of aggregated values for the whole market segment. In the evaluation part of the research, first the predictive model will be implemented and verified using some artificial data. Next, we will apply the model to historical data to determine its effectiveness. A methodology of train-and-test will be used, whereby only the older data is used for training (creating the model) and the newer data is used for testing. Separately, as part of the evaluation, we will perform detailed software testing on the dashboard to ensure that all required functionality is present and works correctly and reliability.