R 利用机器学习预测未来事件

我有一些像这样的事务数据

 SHOP_ID,  DATE   , DAY, IN_TIME , OUT_TIME
shop007, 2017/5/20, mon, 05:03:38, 05:05:33  
shop0010, 2017/4/13, sat, 08:53:42, 08:53:45  
shop005, 2017/10/25, wed, 03:02:42, 03:04:15  
shop001, 2017/10/5, sun, 19:09:37, 19:11:35  
shop008, 2017/1/19, sat, 14:33:01, 14:35:00  
shop004, 2017/3/13, sun, 02:16:06, 02:17:59  
shop0010, 2016/7/4, thu, 10:25:54, 10:25:59  
shop008, 2016/11/6, sat, 22:52:21, 22:53:49  
shop004, 2016/11/13, tue, 08:30:51, 08:32:04  
shop007, 2016/10/2, wed, 19:28:29, 19:29:48  
shop006, 2017/9/25, mon, 01:11:19, 01:12:12    
shop003, 2017/1/14, mon, 00:43:33, 00:43:53  
shop009, 2017/7/7, fri, 16:35:52, 16:36:54  
shop008, 2017/4/26, tue, 06:31:23, 06:33:10  
shop007, 2016/3/19, fri, 04:46:34, 04:48:04  
shop001, 2016/11/4, mon, 11:16:55, 11:18:22    
shop001, 2017/8/31, sat, 07:07:25, 07:09:16  
shop005, 2017/3/16, mon, 17:17:00, 17:18:47  
shop001, 2017/4/23, fri, 04:35:37, 04:37:24  
shop003, 2016/9/18, thu, 08:53:55, 08:55:35  
shop001, 2016/1/12, sun, 10:25:43, 10:26:09  
shop009, 2017/4/9, mon, 17:44:45, 17:45:54  
shop004, 2017/7/1, mon, 01:23:14, 01:24:37  
shop002, 2017/12/28, thu, 18:00:34, 18:00:50  
shop009, 2016/4/6, tue, 00:48:25, 00:49:50  
shop009, 2016/4/10, sat, 14:21:41, 14:22:19  
shop001, 2016/5/16, wed, 15:07:17, 15:09:14  
shop005, 2016/10/6, wed, 23:09:58, 23:10:07  
shop009, 2016/5/6, tue, 09:39:47, 09:39:55  
shop002, 2017/6/16, sat, 19:35:08, 19:35:53  
shop005, 2017/5/26, wed, 10:08:24, 10:09:31  
shop003, 2016/8/7, fri, 06:52:28, 06:52:54  
shop006, 2017/5/5, thu, 17:28:06, 17:28:50  
shop001, 2016/1/7, wed, 10:39:07, 10:39:24  

我想创建一个时间序列模型，利用收集到的语料库数据预测当前一周、一天、一小时内走进的客户数量

我想要的模型是ncustomers~时间，其中ncustomers是客户总数，时间可以是周、日和小时

---

我不知道我们是否可以使用线性回归模型，因为自变量是分类类型，因变量是连续类型

首先，你必须准确定义你的时空。如果你想要一个每日模型，你必须使用日期作为你的可变时间。对于这种情况，我认为您需要的是一个时间序列预测模型

线性回归是一种选择，但在这种情况下，有更多的定性和有用的模型。我给你留了一些

下一步是选择最佳模型，因此您必须采用最小MSE的模型，例如，还有其他度量。我给你一个关于这个嘟嘟的循环

if (!require("forecast")){install.packages("forecast"); library(forecast)} 

train<-ts(your_data_train,frequency=24)
test<-ts(your_data_test,frequency=24)


fit1<-tslm(train ~ trend + season) #linear regression
fcast1 <- forecast(fit1,h=npred ,level = 0)

fit2<-ets(train,ic="aic") #exponencial smoothing without Box-Cox
fcast2 <- forecast(fit2,h=npred ,level = 0)

fit3<-HoltWinters(train) #Clasic Holtwinter (similar to exponencial smoothing)
fcast3 <-forecast(fit3,h=npred ,level = 0)

fit4<-HoltWinters(train,seasonal = "mult") #Holtwinter mult
fcast4 <-forecast(fit4,h=npred ,level = 0)

fit5 <- tbats(train) #Model TBATS
fcast5 <- forecast(fit5, h=npred ,level = 0)

lam <- BoxCox.lambda(train)
fit6 <- ets(train, additive=TRUE, lambda=lam) #Model exponencial smoothing with de Box.Cox
fcast6 <-forecast(fit6,h=npred,level = 0)

pred<-c(fcast1$mean,fcast2$mean,fcast3$mean,fcast4$mean,fcast5$mean,fcast6$mean)

error<-0
for(i in 1:length(pred)){
  error[i]<-mean((pred[i]-test)^2)
}

---

您可以将分类变量映射为一组二进制变量，每个二进制变量仅表示X属于Y=1或X不属于Y=0。这样的二元变量可以作为自变量输入到许多不同的模型中。我们将花一些时间来讲解这些概念。谢谢你的详细回答，非常感谢。我会带着结果回来的：如果你喜欢，请给我一张赞成票，这会对我有很大帮助