基于R语言随机森林模型的事件工单优先级分类实践问询

Hey there! I see you're just getting started with R and machine learning, working on predicting incident ticket priorities from their descriptions—great project! Let's build out your workflow step by step, starting from where you left off.

Your Current Progress #

You've already loaded the key libraries and your dataset, which is a perfect starting point:

library(tm)
library(SnowballC)
library(caTools)
library(randomForest)

incidents = read.csv("incident.csv", stringsAsFactors = FALSE)
str(incidents)

Partial output you shared:
'data.frame': 4265 obs. of 7 variables:
$ number : chr "INC0031193" "INC0037867" "INC0159979" "INC0031446" ...
[remaining variables omitted]

Now let's walk through the critical next steps to turn that raw text data into a working predictive model.

Step 1: Text Preprocessing (Make Text Machine-Friendly) #

Since your predictor is ticket description text, we need to clean and structure it into numerical features that the random forest can use. Here's how to do that:

# Replace "description" with the actual name of your text column in the dataset
corpus = VCorpus(VectorSource(incidents$description))

# Convert all text to lowercase
corpus = tm_map(corpus, content_transformer(tolower))

# Remove punctuation marks
corpus = tm_map(corpus, removePunctuation)

# Strip out common stopwords (e.g., "the", "and") that don't add predictive value
corpus = tm_map(corpus, removeWords, stopwords("english"))

# Stem words (reduce to root form, e.g., "running" → "run")
corpus = tm_map(corpus, stemDocument)

# Convert cleaned text to a Document-Term Matrix (text → numerical features)
dtm = DocumentTermMatrix(corpus)

# Remove rare words to avoid overfitting (keeps terms appearing in at least 1% of tickets)
dtm_sparse = removeSparseTerms(dtm, 0.99)

# Convert to a data frame compatible with modeling
text_features = as.data.frame(as.matrix(dtm_sparse))

# Add your target variable (priority) back to the feature set
text_features$priority = incidents$priority

Step 2: Prep Data for Training #

Random Forest requires categorical target variables to be factors, so let's fix that and split our data into training/test sets:

# Convert priority to a factor (critical for classification)
text_features$priority = as.factor(text_features$priority)

# Set seed for reproducible results
set.seed(123)
# Split data: 70% training, 30% testing
split = sample.split(text_features$priority, SplitRatio = 0.7)
train_data = subset(text_features, split == TRUE)
test_data = subset(text_features, split == FALSE)

Step 3: Train the Random Forest Model #

Now we can build and train our classifier:

set.seed(456)
# ntree = number of trees in the forest (adjust based on performance)
rf_model = randomForest(priority ~ ., data = train_data, ntree = 200)

# Check model summary to see initial performance
print(rf_model)

Step 4: Evaluate Model Performance #

Let's test how well our model works on unseen data:

# Generate predictions on the test set
predictions = predict(rf_model, newdata = test_data)

# Create a confusion matrix to see class-wise performance
confusion_matrix = table(test_data$priority, predictions)
print(confusion_matrix)

# Calculate overall accuracy
accuracy = sum(diag(confusion_matrix)) / sum(confusion_matrix)
cat("Model Accuracy:", round(accuracy * 100, 2), "%\n")

Quick Tips to Boost Performance #

• Tune Hyperparameters: Use tuneRF() to find the optimal mtry (number of variables sampled per tree split) for better results.
• Fix Imbalanced Data: If some priority classes are underrepresented, try oversampling minority classes or using the classwt parameter in randomForest() to assign weights.
• Add More Features: Include other relevant columns (like ticket category, submitter department) alongside text features to give the model more context.

内容的提问来源于stack exchange，提问作者Sourav94