A Parallel Algorithm for Mining Interactive Features from Large Scale Sequences
-
摘要: 序列是一种重要的数据类型，在诸多应用领域广泛存在.基于序列的特征选择具有广阔的现实应用场景.交互特征是指一组整体具有显著强于单独个体与目标相关性的特征集合.从大规模序列中挖掘交互特征面临着位点的“组合爆炸”问题，计算挑战性极大.针对该问题，以生物领域高通量测序数据为背景，提出了一种新的基于并行处理和演化计算的高阶交互特征挖掘算法.位点数是制约交互作用挖掘效率的根本因素.摈弃了现有方法基于序列分块的并行策略，采用基于位点分块的并行思想，具有天然的效率优势.进一步，提出了极大等位公共子序列(maximal allelic common subsequence, MACS)的概念并设计了基于MACS的特征区域划分策略.该策略能将交互特征的查找范围缩小至许多“碎片”空间，并保证不同“碎片”间不存在交互特征，避免计算耦合引起的高额通信代价.利用基于置换搜索的并行蚁群算法，执行交互特征选择.大量真实数据集和合成数据集上的实验结果，证实提出的PACOIFS算法在有效性和效率上优于同类其他算法.Abstract: Sequence is an important type of data which is widely existing in various domains, and thus feature selection from sequence data is of practical significance in extensive applications. Interactive features refer to a set of features, each of which is weakly correlated with the target, but the whole of which is strongly correlated with the target. It is of great challenge to mine interactive features from large scale sequence data for the combinatorial explosion problem of loci. To address the problem, against the background of high-throughput sequencing in biology, a parallel evolutionary algorithm for high-order interactive features mining is proposed in this paper. Instead of sequence-block based parallel strategy, the work is inspired by loci-based idea since the number of loci is the fundamental factor that restricts the efficiency. Further, we propose the conception of maximal allelic common subsequence (MACS) and MACS based strategy for feature region partition. According to the strategy, the search range of interactive features is narrowed to many fragged spaces and interactions are guaranteed not to exist among different fragments. Finally, a parallel ant algorithm based on substitution search is developed to conduct interactive feature selection. Extensive experiments on real and synthetic datasets show that the efficiency and effectiveness of the proposed PACOIFS algorithm is superior to that of competitive algorithms.
-
-
期刊类型引用(9)
1. 段雪莹，王立君. 有向复杂网络软件异常交互执行行为挖掘算法. 计算机仿真. 2023(01): 533-538 . 
百度学术
2. 许中平，赵恩来，张鹤译，牟玮，丁玉星. 基于知识图谱的智能电网多维数据关联挖掘方法. 电子设计工程. 2023(11): 84-87+92 . 百度学术
3. 唐明康，王科盛，李双双，刘培，彭旭光. 基于物联网数据和神经网络的呼吸机故障预测方法研究. 医疗卫生装备. 2023(09): 8-13 . 百度学术
4. 周翔，翟俊海，黄雅婕，申瑞彩，侯璎真. 大数据环境下的投票特征选择算法. 小型微型计算机系统. 2022(05): 936-942 . 百度学术
5. 王逸兮，廖荣涛，叶宇轩. 基于中台架构的内生性数据安全交互系统设计. 电子设计工程. 2022(12): 153-157 . 百度学术
6. 王赫楠，孙艳秋，张柯欣. 数据集压缩建模的研究. 计算机与数字工程. 2022(06): 1286-1291+1364 . 百度学术
7. 蔡梦睿. 基于数据挖掘的网络隐私信息安全加密系统. 信息技术. 2022(09): 94-99 . 百度学术
8. 张辽艳. 基于特征挖掘的相似音乐片段检测方法研究. 自动化技术与应用. 2022(10): 152-154+170 . 百度学术
9. 韩阳，王文娟，李俊. 基于虚拟桌面终端技术的船舶通信数据交互方法. 舰船科学技术. 2020(20): 145-147 . 百度学术
其他类型引用(3)
计量
- 文章访问数: 1208
- HTML全文浏览量: 0
- PDF下载量: 429
- 被引次数: 12
下载:
