本文主要采用了文献与理论研究、实证研究两种研究方法，具体如下:(1)文献与理论研究阅读国内外量化选股相关文献，了解量化选股常用方法，归纳研究现状:阅读机器学习排序相关文献，总结机器学习排序的原理以及相关算法:阅读模式识别、动量反转相关的选股理论与方法，分析归纳传统方法的不足，以构建基于机器学习的模式识别、动量反转选股策略。

(2)实证研究方法选取覆盖牛、熊和震荡行情的股价数据作为训练集和测试集，使用python进行编程和数据处理，从而对策略的有效性进行检验。

在特征提取中，采用ema降噪、z-score 标准化、分位数化等数据处理方法对特征向量进行处理，以提高机器学习模型的泛化能力。

4. 参考文献

[1]丁鹏.量化投资与对冲基金入门(量化投资与对冲基金从书)[M].电子工业[2] Bischi G I, Valori V. Nonlinear effects in a discrete time dynamic model ofa stockmarket[J]. Chaos Solitons Fractals[3] Rai P, Prabhumoye s, Khatri P, et al. A Prototype of an Iellien Search EngineUsing Machine Learning Based Training for Learning to Rank[J]. SmartInnovation Systems Technologies[4] Dror G Koren Y, Maarck Y, et al. I want to answer, who has a question?: Yahoo!answers recommender system.[]. Proceedings of Kdd[5] Thomas S, Seltzer M L, Church K, et al. Deep neural network features and scmi-supervised training for low resource speech recognition[C]/ Acoustics,Speech, and Signal Processing, 1988.ICASSP-88.. 1988 Intermational Conferenceon.[6] Liu Y, Zhang Y, Che W, et al. Domain adaptation for crf-based Chincse wordsegmentation using free annotations(CV/Proceedings of the 2014 Conference onEmpirical Methods in Natural Language Processing. 2014: 864-874.[7] Fan A, Palaniswami M. Stock selection using Support Vector Machines[C]/l Neural Networks, 2001. Proceedings. IJCNN 01. Intemational Joint Conference on. IEEE,2001:1793- 1798.[8]Li B, Zhao P, Hoi S C H, et L PAMR: Passive aggressive mean reversion strategy for porfolio selection[J]- Machine Leaming, 2012, 87(2):221-258.[9]Li B, Hoi S C H, Zhao P, et al. Confidence Weighted Mean Reversion Strategy for Online Porfolio Selection[J]. Acm Transactions on Knowledgc Discovery fromData, 2013, 7(1):82-83.[10]阿基利斯技术分析指标大全[M].机械工业出版社，2011.