Forecasting Study In Specific Data Situation

With the fast development of computer science and digital technology,all human’s behaviors have been recorded by digital equipment. The big data and the cloud computing have set off a wave of academic research. Then, it is well known that the one of a core problem of big data is the prediction. The research of prediction can help forecast the future and make decisions efficiently. However, we should pay attention to differences existing in these established models and suitable methods should be chosen to make up the model.In this research, the author builds up two prediction models with big data and small data separately. The first one is the sales market prediction model in fast fashion industry. Different from the previous research, this paper aims to forecast the market trend of such products. The sales data acquired from the historical time series is counted weekly and be cut into three market types. So the model is built from a multi-class classification point of view. However, since the fast fashion products own the features of fast updating and short sales cycles, the data for a typical neural network is hardly to be used to train the model. Based on this phenomenon, this paper chooses a newly proposed method—the extreme learning machine which owns the features of fast learning speed, simple structure, and has been widely used in forecasting and classification problems. The second model is the stock forecast with complex data structure since the stock price is determined by many different factors. Based on the features of the training data, the author designed many variables which contain all the main affecting factors and chose the popular deep learning methods to extract important information of the training data and make learning more efficient. For all these models, other existing popular techniques are token to make comparisons. By extensive experiments, the author demonstrates that the choosing methods are very suitable to our forecasting models.