关于数据挖掘关联规则的Oracle实现

呵呵，前几天拿到了数据挖掘基础教程一书，感觉部分算法是基于统计学的原理的，而统计学是可以通过Oracle来实现。

关于数据挖掘关联规则的介绍，可以参见：http://baike.baidu.com/view/1076817.htm?fr=ala0_1

关联规则是形如X→Y的蕴涵式，

其中且， X和Y分别称为关联规则的先导(antecedent或left-hand-side, LHS)和后继(consequent或 right-hand-side, RHS) 。

关联规则在D中的支持度(support)是D中事务同时包含X、Y的百分比，即概率；=X^Y/D

置信度(confidence)是包含X的事务中同时又包含Y的百分比，即条件概率。 =(X^Y)/X

关联规则是有趣的，如果满足最小支持度阈值和最小置信度阈值。

若给定最小支持度α = n，最小置信度β = m，则分别通过以上的X^Y/D和(X^Y)/X，可获知是否存在关联

使用的原始数据

反范式后的数据

待统计项

代码示例

代码语言：javascript复制

--创建各个购买单元项视图
create view distinct_trans as select distinct tranobject from purchase;

--创建各个事务内部的购买单元项
--可以用wm_concat函数
create view all_trans as
SELECT tranid,MAX(tranobjects) tranobjects
  FROM (SELECT tranid,WMSYS.WM_CONCAT(tranobject) OVER(PARTITION BY tranid ORDER BY tranobject) tranobjects
               FROM purchase)
GROUP BY tranid;

--也可以用sys_connect_by_path函数
create view all_trans as
select tranid,substr(tranobjects,2) tranobjects
  from  --格式化前面的逗号和空格
  (select distinct tranid,FIRST_VALUE(tranobjects) OVER(PARTITION BY tranid ORDER BY levels desc ) AS tranobjects  --保留最大的那个
     from
     (select tranid,sys_connect_by_path(tranobject,',') tranobjects,level levels --各购买事务的内部排列组合
        from purchase
      connect by tranid=prior tranid and tranobject
     )
  );


--对所有购买单元项进行排列组合，即数据挖掘的X^Y项
create view all_zuhe as
select substr(sys_connect_by_path(tranobject,','),2) zuhe
  from (select distinct tranobject from purchase)
connect by nocycle tranobject

select * from all_zuhe

--筛选出符合要求的排列组合，即数据挖掘的X项和Y项
create view full_zuhe as
select a.zuhe X,b.zuhe Y from all_zuhe a,all_zuhe b
 where instr(a.zuhe,b.zuhe)=0 and instr(b.zuhe,a.zuhe)=0
   and not exists(select 1 from distinct_trans c
                          where instr(a.zuhe,c.tranobject)>0 and instr(b.zuhe,c.tranobject)>0)


select * from full_zuhe

create or replace view tongji as
select xy,xy_total,x,x_total,y,y_total,transtotal
  from
  (select y||','||x xy,
         (select count(*) from all_trans a where instr(a.tranobjects,c.x||','||c.y)>0 or instr(a.tranobjects,c.y||','||c.x)>0) xy_total, --包含xy的事务数
         y,
         (select count(*) from all_trans b where instr(b.tranobjects,c.y)>0) y_total, --包含y的事务数
         x,
         (select count(*) from all_trans b where instr(b.tranobjects,c.x)>0) x_total, --包含x的事务数
         d.transtotal  --总事务数
   from full_zuhe c,(select count(distinct tranid) transtotal from purchase) d
  order by xy_total desc,x_total desc
)

select * from tongji where xy_total>=3 and y_total>=3

数据挖掘

0 人点赞