基因表达微阵列数据分类的多目标启发式算法

引用

LaTex

@article{LV201613,
title = “A multi-objective heuristic algorithm for gene expression microarray data classification”,
journal = “Expert Systems with Applications”,
volume = “59”,
pages = “13 - 19”,
year = “2016”,
issn = “0957-4174”,
doi = “https://doi.org/10.1016/j.eswa.2016.04.020“,
url = “http://www.sciencedirect.com/science/article/pii/S0957417416301865“,
author = “Jia Lv and Qinke Peng and Xiao Chen and Zhi Sun”,
keywords = “Microarray, Gene selection, Small number of selected genes, Multi-objective, Heuristic algorithm”
}

Normal

Jia Lv, Qinke Peng, Xiao Chen, Zhi Sun,
A multi-objective heuristic algorithm for gene expression microarray data classification,
Expert Systems with Applications,
Volume 59,
2016,
Pages 13-19,
ISSN 0957-4174,
https://doi.org/10.1016/j.eswa.2016.04.020.
(http://www.sciencedirect.com/science/article/pii/S0957417416301865)
Keywords: Microarray; Gene selection; Small number of selected genes; Multi-objective; Heuristic algorithm

---

摘要

Microarray data 微阵列数据
analytic hierarchy process (AHP)
Univariate Marginal Distribution Algorithm

the fewer the selected genes are, the less cost the disease prognosis expert system is.

---

主要内容

---

1 特征预选择

a filter-based gene ranking algorithm — mRMR：
特征与类之间的相关性（max-relevance 最大相关）
特征之间的冗余度（min-redundancy 最小冗余）

单个特征的性能
为防止丢失在组中表现好的特征，选300个特征

---

2 多目标模型

---

3 MOEDA

多目标the estimation of distribution algorithm (EDA) — MOEDA

elite individuals ( EIs )
regenerated individuals ( RIs )

probabilistic model：

classification accuracy (ACC)
the number of selected features (NSF)

Higher and fewer rule. (HFR)
ACC绝对比NSF重要
- 根据ACC对个体排序
- 对于相同ACC，根据NSF排序

Forcibly decrease rule. (FDR)
随着演化的进行，计算NSF的上限 —$UL^l$（逐渐降低）
$NL^l = \frac{q}{2^{\left\lfloor\frac{l}{w}\right\rfloor}}$
$l$ — 代数
$q$ — 预选择的特征数目
$w$ — 常数

每个特征对应一个选择概率

mutation rules — 防止落入局部最优
the elite reserved strategy — 防止最优个体丢失

SVM + the radial basis function (RBF)
SVM-RBF
参数：$c$与$\gamma$

同时优化参数与特征

参数计算

$p \in \left\{ c, \gamma \right\}$
$\max_p$ — 参数最大值
$\min_p$ — 参数最小值
$d$ — 二进制字符串的十进制值
$l_p$ — 二进制字符串的长度
$l_c = l_\gamma = 25$
$\max_c = 256$
$\max_\gamma = 16$

---

4 试验

10-fold cross validation

‘the N best features are always not the best N features’.