A Self-adaptive Weighted SVM Stock Trend Forecasting Model

With the continuous development and improvement of the securities market,the stock market as a function of China's economic barometer has become increasingly prominent.Both the individual investors and the country are very concerned about the trend of the stock market.If the trend of the stock market can be accurately predicted,it can not only provide a basis for investors to make investment decisions,but also provide a reference for the country to formulate relevant economic policies.The stock market has characteristics such as non-linearity,high noise and large data volume.The related stock analysis methods,such as fundamental analysis method,technical analysis method and time series analysis method,have their own characteristics,but they are difficult to adapt to the increasingly complex stock market.Research shows that the accuracy of stock trend forecasting is mainly affected by both data quality and algorithm performance.Therefore,this thesis proposes a self-adaptive weighted SVM stock trend forecasting model,and designs two sub-algorithms.One is based on genetic algorithm(GA)adaptively weighting data features,and the other is using particle swarm optimization(PSO)to optimize SVM parameters.Then,14 key stock financial indicators are selected.Finally,the stock trend forecasting model is implemented.The experimental results on the actual stock data set show that the model proposed in this thesis is feasible and effective,and its performance is better than Decision Tree model,KNN model,Bayesian model and BPNN model.The main work of this thesis is as follows:(1)According to the characteristics of stock data,this thesis proposes a stock trend prediction model based on GA feature adaptive weighting and PSO to optimize SVM parameters,namely PSO-GA-SVM model.(2)In the aspect of improving input data quality,the GA is used to adaptively weight the characteristics of the stock data in order to highlight important attributes,constrain the redundancy or secondary attributes,and further show the influence of different attributes on stock prices.(3)In the aspect of algorithm performance improvement,because the performance of SVM is very sensitive to its parameters,PSO is used to optimize the parameters of SVM,such as penalty parameter c and kernel function parameter g,so as to solve the problem of SVM parameter selection.(4)Selection of stock price indicators.On the basis of summarizing previous research results,14 financial indicators affecting stock prices were selected from 6 aspects: profitability,growth capacity,liquidity,rationality of stock prices,efficiency,and financial leverage.