Search code examples
rglm

predict some missing values based on linear model

enter image description here

datax <- matrix(1:32, nrow = 8)
datax[2:5,1] <- NA
m <- data.frame(datax)
names(m)[c(1:4)] <- c("Length", "Width", "sex", "height")
model <- glm(Length ~ Width + sex + height, data = m)

How do you predict the NA values based on the model (code just given as example)

I've got a dataset with 15 variables and the response variable has some missing values. How can I predict the missing values of the response variable based on a linear model built from this dataset?

Solution

  • How about subsetting your data into parts with and without missing values, creating a linear model based on the latter and imputing the missing values on the former through predict()?

    library(tidyverse)

datax <- matrix(1:32, nrow = 8)
datax[2:5,1] <- NA
m <- data.frame(datax)
names(m)[c(1:4)] <- c("Length", "Width", "sex", "height")

# Creating an index of rows with missing values in "Length"
missing_index <- which(is.na(m$Length))

# Subsetting rows with missing values
m_missing <- m[missing_index,]

# Subsetting the rest
m_rest <- m[-missing_index,]

# Creating a linear model on m_rest and making predictions on m_missing
model <- lm(Length ~ ., data = m_rest)
predictions <- predict(model, newdata = m_missing %>% select(-Length))

# Insert missing values into the original dataframe
m[missing_index, "Length"] <- predictions

    Resulting in:

    > print(m)
  Length Width sex height
1      1     9  17     25
2      2    10  18     26
3      3    11  19     27
4      4    12  20     28
5      5    13  21     29
6      6    14  22     30
7      7    15  23     31
8      8    16  24     32