留存率预测模型

留存率预测模型

由资料和模型可知：留存率曲线是一个指数曲线，可根据前7天留存率数据预测之后的留存率数据：利用nls函数求出幂指数函数y=a*x^b的系数a、b

# 前七天实际留存率数据
(day <- seq(1:7)) # 天数
(ratio <- c(0.383,0.268,0.216,0.187,0.167,0.156,0.145)) # 留存率值
# 利用nls函数求出幂指数函数y=a*x^b的系数a、b
fit <- nls(ratio~a*day^b,start = list(a=1,b=1))
# 查看模型结果
summary(fit)
# 对新增用户在接下来365日每天的留存率进行预测
predicted <- predict(fit,data.frame(day=seq(1,365)))
# 查看预测结果
predicted
# 绘制留存率预测曲线
library(dygraphs)
data <- as.data.frame(predicted)
data <- ts(data)
dygraph(data,main="留存的预测曲线") %>% 
  dySeries("predicted",label="留存率",strokeWidth = 2) %>%
  dyOptions(colors = "green",fillGraph = TRUE,fillAlpha = 0.4) %>%
  dyHighlight(highlightCircleSize = 5, 
              highlightSeriesBackgroundAlpha = 0.2,
              hideOnMouseOut = FALSE) %>%
  dyAxis("x", label = "日期",drawGrid = FALSE) %>% 
  dyAxis("y", label = "留存率") %>%
  dyRangeSelector()

结果如下：

> (day <- seq(1:7)) # 天数
[1] 1 2 3 4 5 6 7
> (ratio <- c(0.383,0.268,0.216,0.187,0.167,0.156,0.145)) # 留存率值
[1] 0.383 0.268 0.216 0.187 0.167 0.156 0.145

#参数估计结果
> summary(fit)

Formula: ratio ~ a * day^b

Parameters:
   Estimate Std. Error t value Pr(>|t|)    
a  0.381911   0.002164  176.51 1.11e-10 ***
b -0.508544   0.005571  -91.29 2.99e-09 ***
---
Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

Residual standard error: 0.002349 on 5 degrees of freedom

Number of iterations to convergence: 7 
Achieved convergence tolerance: 9.236e-08


#预测留存率结果：
> predicted
  [1] 0.38191061 0.26845690 0.21843606 0.18870675 0.16846294 0.15354554
  [7] 0.14196843 0.13264787 0.12493582 0.11841787 0.11281510 0.10793196

可放大图形，看更细致的曲线：