金融时间序列预测对经济决策和投资意义重大,但金融市场的复杂性给预测模型构建带来挑战,而黄金价格走势备受关注,准确预测至关重要。本文针对现有组合模型不足,提出创新的非线性ARIMA-LSTM组合模型用于黄金价格预测。实证分析发现,ARIMA(3,1,5)模型、LSTM模型及GRU模型虽能捕捉时间序列特征但预测存在偏差,结果表明组合模型ARIMA-LSTM预测效果优于其他三种模型。通过MAE和RMSE评估,验证了ARIMA-LSTM模型在黄金价格预测中的优势,为金融决策提供新思路。Financial time series forecasting is of great significance to economic decision-making and investment, but the complexity of financial markets brings challenges to the construction of forecasting models, and the trend of gold price has attracted much attention, so accurate forecasting is crucial. This paper aims at the shortcomings of existing combination models, an innovative nonlinear ARIMA-LSTM combined model is proposed for gold price prediction. The empirical analysis shows that although ARIMA(3,1,5) model, LSTM model and GRU model can capture the features of time series, the prediction bias exists. The results show that the combined model ARIMA-LSTM has better prediction effect than the other three models. Through MAE and RMSE evaluation, the advantages of ARIMA-LSTM model in gold price prediction are verified, which provides new ideas for financial decision-making.
有效而准确的预测商品混凝土价格变动趋势,对各类建筑的施工规划具有重要意义。相比其他预测模型,随机森林模型具有更高的预测精度。然而不同的数据结构都有其独特之处,针对特定数据结构进行模型优化,有助于提高算法在特定数据上的处理性能。我们针对时间序列分类(TSC:Time Series Classification)的特征提出一种改进随机森林算法。首先将随机森林创建训练子集时的随机抽样调整为倾斜抽样,然后将决策树分裂时的随机特征向量抽样调整为分层抽样,最后以加权投票取代平均投票。实证结果表明相比原始随机森林算法,改进模型具有明显优势,对商品混凝土价格变动的预测准确率达98.4%,预测精度、召回率和F1评分分别为:98.7%,98.2%,98.4%,可以实现了商品混凝土价格变动趋势的精准预测。