Classification and Clustering

library(rpart)
library(rpart.plot)
library(caret)

set.seed(42)
# Customer loan default prediction
n <- 600
loans <- data.frame(
  income      = round(runif(n, 25000, 150000)),
  debt_ratio  = round(runif(n, 0.1, 0.8), 2),
  age         = sample(21:65, n, replace=TRUE),
  loan_amount = round(runif(n, 5000, 50000)),
  credit_score= sample(450:800, n, replace=TRUE)
)
loans$default <- factor(
  ifelse(loans$debt_ratio > 0.5 & loans$credit_score < 600, "Yes",
         ifelse(loans$income < 40000 & loans$debt_ratio > 0.4, "Yes", "No")),
  levels=c("No","Yes")
)

# Train/test split
train_idx <- createDataPartition(loans$default, p=0.8, list=FALSE)
train <- loans[train_idx, ]
test  <- loans[-train_idx, ]

# Decision Tree
tree_model <- rpart(default ~ ., data=train, method="class",
                     control=rpart.control(maxdepth=5, minsplit=20))

# Visualize tree
rpart.plot(tree_model, type=4, extra=104, fallen.leaves=TRUE,
            main="Loan Default Decision Tree")

# Evaluate
pred_tree <- predict(tree_model, newdata=test, type="class")
cm <- confusionMatrix(pred_tree, test$default, positive="Yes")
cat(sprintf("Decision Tree: Accuracy=%.3f, F1=%.3f\n",
             cm$overall["Accuracy"], cm$byClass["F1"]))

library(class)

# Standardize features
train_x <- scale(train[, -ncol(train)])
test_x  <- scale(test[, -ncol(test)],
                  center=attr(train_x,"scaled:center"),
                  scale=attr(train_x,"scaled:scale"))

# Find optimal k using cross-validation
k_values <- 1:20
errors <- sapply(k_values, function(k) {
  pred <- knn(train_x, test_x, train$default, k=k)
  mean(pred != test$default)
})
optimal_k <- k_values[which.min(errors)]
cat(sprintf("Optimal k: %d (error: %.3f)\n", optimal_k, min(errors)))

# Final KNN model with optimal k
pred_knn <- knn(train_x, test_x, train$default, k=optimal_k, prob=TRUE)
cm_knn <- confusionMatrix(pred_knn, test$default, positive="Yes")
cat(sprintf("KNN (k=%d): Accuracy=%.3f, F1=%.3f\n",
             optimal_k, cm_knn$overall["Accuracy"], cm_knn$byClass["F1"]))

library(cluster)
library(factoextra)

# ─── CUSTOMER SEGMENTATION SYSTEM ────────────────────
set.seed(42)
n <- 400
customers <- data.frame(
  customer_id    = 1:n,
  annual_spend   = round(runif(n, 500, 10000), -2),
  visit_frequency= sample(1:50, n, replace=TRUE),
  avg_basket     = round(runif(n, 20, 200), -1),
  tenure_months  = sample(3:72, n, replace=TRUE),
  returns_pct    = round(runif(n, 0, 0.3), 2)
)

# Feature selection and scaling
features <- customers %>% select(annual_spend, visit_frequency, avg_basket, tenure_months)
features_scaled <- scale(features)

# Find optimal clusters (elbow method)
wss <- sapply(2:10, function(k) {
  kmeans(features_scaled, centers=k, nstart=25)$tot.withinss
})
# Plot elbow
plot(2:10, wss, type="b", pch=19, main="Elbow Method",
     xlab="Number of Clusters", ylab="Total Within SS")

# Fit k-means with k=4
set.seed(42)
km <- kmeans(features_scaled, centers=4, nstart=25)
customers$cluster <- factor(km$cluster)

# Cluster profiles
cluster_profile <- customers %>%
  group_by(cluster) %>%
  summarise(
    n            = n(),
    avg_spend    = round(mean(annual_spend)),
    avg_visits   = round(mean(visit_frequency), 1),
    avg_basket   = round(mean(avg_basket)),
    avg_tenure   = round(mean(tenure_months)),
    .groups = "drop"
  ) %>%
  arrange(desc(avg_spend))

cat("=== CUSTOMER SEGMENTS ===\n")
print(cluster_profile)

# Name the segments
cluster_names <- c("1"="Champions", "2"="Loyal Customers",
                    "3"="At-Risk", "4"="New Customers")
customers$segment <- cluster_names[as.character(customers$cluster)]
cat("\nSegment Distribution:\n")
print(table(customers$segment))

# Visualize clusters
fviz_cluster(km, data=features_scaled, geom="point",
              ellipse.type="convex", ggtheme=theme_minimal(),
              main="Customer Segmentation (4 Clusters)")

# Silhouette score (cluster quality: -1 to 1, higher is better)
sil <- silhouette(km$cluster, dist(features_scaled))
cat(sprintf("\nAverage Silhouette Score: %.3f\n", mean(sil[,3])))