Financial Index Forecast Based On Data Mining

Time series mining is an important research domain of data mining. Financial time series mining is a difficult aspect of time series mining, because its non-stationarity, autocorrelation, non-linear, low signal to noise ratio, etc. But it has high value of research and market at the same time. As to stock data, financial index has good transparency and anti-manipulation, so it is a proper research target.In this paper, we construct a 60-minute high-frequency trading strategy based on the market trading rule and the cycle characteristics. Then we dig out a tradable time units set by data preprocessing, feature extraction and k nearest neighbors classification, which meet the above trading strategy. Cycle characteristics are key consideration in data preprocessing. In feature extraction, the volume ratio is key consideration and weighted volume ratio is proposed. Based on the improvement of the traditional kNN algorithm, the probability kNN algorithm is proposed to adapting financial data. At last, the experiments on actual data with Matlab show that these improvements achieve good results.