基于时间和聚类分组的多变量预测R代码需求

Hey there! As someone who’s worked with R for years and helped plenty of beginners, I’ll walk you through a straightforward, step-by-step solution to build your grouped prediction table and export it to CSV. Let’s keep this simple and actionable since you’re new to R.

First, we’ll use the tidyverse package—it’s a beginner-friendly toolkit that makes grouping data, running predictions, and exporting files way easier. Install and load it like this:

# Install the package if you haven't already
install.packages("tidyverse")

# Load it into your R session
library(tidyverse)

Let’s make a test dataset that mirrors your example. This lets you test the code before using your real data:

set.seed(123) # Ensures random numbers are reproducible

sample_data <- tibble(
  Time = rep(seq.Date(as.Date("2018-04-21"), as.Date("2018-04-25"), by = "day"), 2),
  Cluster = rep(c("A", "B"), each = 5),
  X1 = sample(10:80, 10, replace = TRUE),
  X2 = sample(30:90, 10, replace = TRUE),
  X3 = sample(20:80, 10, replace = TRUE)
)

print(sample_data)

We’ll build a simple function that takes a single Cluster’s time-series data and generates predictions. For this example, we’ll use a lag-based prediction (predict the next day’s values using the most recent observed values)—you can swap this for a more complex model (like linear regression) later if needed:

# Function to predict the next time step for a single Cluster
predict_group <- function(group_data) {
  # Get the last recorded date and calculate the next date
  last_time <- max(group_data$Time)
  next_time <- last_time + days(1)
  
  # Generate predictions using the latest observed values
  predictions <- group_data %>%
    slice_tail(n = 1) %>% # Grab the most recent row
    mutate(
      Time = next_time,
      X1_pred = X1, # Replace with model output if using regression
      X2_pred = X2,
      X3_pred = X3
    ) %>%
    select(Time, Cluster, X1_pred, X2_pred, X3_pred)
  
  return(predictions)
}

*Pro tip: If you want to use a regression model instead, here’s a quick snippet for X1:

model_x1 <- lm(X1 ~ Time, data = group_data)
x1_pred <- predict(model_x1, newdata = tibble(Time = next_time))
```*

# Step 4: Apply Predictions to Every Cluster
Now we’ll group the data by `Cluster`, run the prediction function on each group, and combine all results into one table:
```r
# Group by Cluster, run predictions, and flatten the results
prediction_results <- sample_data %>%
  group_by(Cluster) %>%
  nest() %>% # Nest each Cluster's data into a separate table
  mutate(predictions = map(data, predict_group)) %>% # Run predictions on each group
  unnest(predictions) %>% # Turn nested predictions into a flat table
  select(-data) # Remove the unused nested data column

# Optional: Combine original data with predictions
combined_data <- sample_data %>%
  bind_rows(prediction_results) %>%
  arrange(Time, Cluster)

print(combined_data)

Finally, export your table (either just predictions or combined original + predictions) to a CSV file you can use later:

# Export combined data to CSV
write_csv(combined_data, "prediction_table.csv")

# Or export only the predictions:
# write_csv(prediction_results, "only_predictions.csv")

• Predict multiple time steps: Add a loop inside predict_group() to generate predictions for 2+ future days.
• Use your real data: Replace sample_data with your actual dataset (load it with read_csv("your_real_data.csv")).
• Add more variables: Extend the prediction logic to include X4, X5, etc.—just mirror the pattern for X1/X2/X3.

内容的提问来源于stack exchange，提问作者B. Alt