This package is designed to support classification & regression modeling pipelines

✨ Install from GitHub

install.packages("devtools")
devtools::install_github("xxxxxi0001/xxml")

✨ Author E-mail

zhang.xi6@northeastern.edu

✨ Features

One Step Data Cleaning

• automate_data_cleaning — One step outlier & NA treatment based on distribution & portion of NA

Data Cleaning Utilities

• check_na_zero() — Detect NA / Zero in all numeric variables

• replace_zero_with_na() — Convert designated zero values to NA

• median_imputation() — Impute NA using median

• z_score_outlier() — Replace outliers using z-score capping

• IQR_outlier() — IQR capping for skewed variables

• check_multicollinearity() — Detect high correlations

• automation_knn_imputation() — one-step knn imputation

  • initialize_distance_find_best_k()
  • initialize_not_na_index()
  • initialize_test_k_index()
  • find_best_k()
  • kNN_Imputation()

(Categorical: Support rf, c5, rpart, lr)

Data Splitting

• three_set_partition() — stratified train/test/validation split
• ensemble_train_partition() — partitions for bag ensemble

Class Imbalance

• check_class_imbalance() - check if class imbalance occur

Cross Validation

• k_fold_stratified() - stratified train index into k fold for pipeline testing
• cross_validation() - cross validation support rf, c5, rpart, lr

Logistic Regression (Backward p-selection)

• backward_p_lr() — iterative removal of non-significant predictors

Ensemble Learning

• make_ensemble_predict_categorical() — mean probability ensemble
• find_best_threshold() — threshold tuning for F1
• ensemble_weight_F1() — model weights from F1 performance
• emsemble_result_with_weight_categorical() — weighted ensemble prediction

Check Modle Performance

• check_model_performance() - TPR, TNR, F1, Accuracy

Stack (categorical)

• generate_stack_df() - helper func
• stack_model_lr() - generate new model with exsiting model with lr
• stack_model_predict_lr() - use stack model make prediction

(Regression)

Data Splitting

• three_set_partition_no_target() — randomly partitioned index into designated portion
• ensemble_train_partition_no_target() — randomly partitioned train index into designated portion

Multi-Linear Regression (Backward p-selection)

• backward_p_mlr() - iterative removal of non-significant predictors

Ensemble Learning

• reverse_num() - if feature changed in feature transformation, change it back
• make_ensemble_predict_numerical() - use ensemble model make mean prediction
• ensemble_weight_with_RMSE() - calculate each model's weight based on RMSE
• ensemble_result_with_weight_numerical() - final prediction with weight

Stack (Regression)

• create_stack_model_mlr() - generate a new multi-linear regression model based on existing model
• stack_test_mlr() - check how stack model perform

---

❤️ Usage Example (Categorical)

library(xxml)

Load your data

df <- read.csv("your_data.csv")

Basic NA / zero check

check_na_zero(df)

Convert zero to NA for Cholesterol

df <- replace_zero_with_na(df, ignore_cols = "Oldpeak")

One-Step Clean Data with Outlier & NA

df <- automate_data_cleaning(df)

Feature Engineering (Derive New Feature, Transformation, Multicolineariality, PCA)

df_LR$doubleRISK<-df_LR$MaxHR * df_LR$Oldpeak


df_LR$Age_square<-(df_LR$Age)^2
hist(df_LR$Age_square,prob=TRUE, main="Histogram of Age After", xlab="Age")
lines(density(df_LR$Age_square, na.rm=TRUE))

variables<-df_LR[,c("Age_square","RestingBP","Cholesterol","MaxHR","Oldpeak_square","doubleRISK_sqrt")]
check_multicollinearity(variables)

pca_results<-prcomp(df_LR[,c("Age_square","RestingBP","Cholesterol","MaxHR","Oldpeak_square","doubleRISK_sqrt")])
summary(pca_results)
pca_df<-as.data.frame(pca_results$x[,1:5])

Partition dataset (50% train, 25% test, 25% val)

index <- three_set_partition(df, "HeartDisease", 0.5, 0.25, 1, 0)

train_index <- index$train_index
test_index  <- index$test_index
val_index   <- index$validation_index

Fit logistic regression backward p

log_models <- backward_p_lr(df, train_index, "HeartDisease", 1, 1, 1)

Ensemble prediction

pred <- make_ensemble_predict(log_models, df, test_index, positive = 1)

Evaluate

check_model_performance(pred, 0.5, 1, 0, df, test_index, "HeartDisease")

Tune with Best Threshold & Weight

best_threshold<-find_best_threshold(ensemble_predictions,df_encoded_LR,test_index,"HeartDisease",1,0)

weight_list<-ensemble_weight_F1(logistic_model_list,df_encoded_LR,test_index,best_threshold,"HeartDisease",1,0)

ensemble_predictions_val<-ensemble_result_with_weight(logistic_model_list,df_encoded_LR,validation_index,weight_list,1)

result_ensemble<-check_model_performance(ensemble_predictions_val,best_threshold,1,0,df_encoded_LR,validation_index,"HeartDisease")

❤️ Usage Example (Regression)

library(xxml)

Load your data

df <- read.csv("your_data.csv")

Basic NA / zero check

check_na_zero(df)

Convert zero to NA for Cholesterol

df <- replace_zero_with_na(df, ignore_cols = "Oldpeak")

One-Step Clean Data with Outlier & NA

df <- automate_data_cleaning(df)

Feature Engineering (Transformation, Multicollinearity, Normalization, One-Hot)

df$DONATION_AMT_log<-log(df$DONATION_AMT)
hist(df$DONATION_AMT_log,prob=TRUE, main="Histogram of Donation Amount After", xlab="Donor Age")
lines(density(df$DONATION_AMT_log, na.rm=TRUE))

variables<-df[,c("DONATION_AMT_log","DONOR_AGE","INCOME_LEVEL","SES","MEDIAN_HOME_VALUE_log","MEDIAN_HOUSEHOLD_INCOME_sqrt","DONATION_RESPONSE_sqrt1","MONTHS_SINCE_LAST_GIFT_square","EMAILS_12_log","LIFETIME_GIFT_COUNT_sqrt","LIFETIME_EMAILS_sqrt","LIFETIME_GIFT_AMOUNT_sqrt","LIFETIME_MAX_GIFT_AMT_sqrt","LIFETIME_MIN_GIFT_AMT_sqrt","LIFETIME_AVG_GIFT_AMT_sqrt")]

check_multicollinearity(variables)

df_scaled[col_numeric]<-scale(df[col_numeric==TRUE])

onehot_features<-model.matrix(~URBANICITY+DONOR_GENDER+HOME_OWNER,data=df_scaled)
onehot_df<-as.data.frame(onehot_features)[,-1]

Partition (50,25,25)

set.seed(888)
partition_result<-three_set_partition_no_target(df_encoded,0.5,0.25)
train_index<-partition_result$train_index
test_index<-partition_result$test_index
validation_index<-partition_result$validation_index

Construction with Backward p

mlr_list<-backward_p_mlr(df_encoded,train_partition_index,"DONATION_AMT_log")

Tune with Weight

weight_list<-ensemble_weight_RMSE(mlr_list,df_encoded,test_index,"DONATION_AMT_log","log")
prediction<-emsemble_result_with_weight(mlr_list,df_encoded,validation_index,weight_list,"log")

Check Result with RMSE & Correlation & Plot

real_value<-exp(df_encoded[["DONATION_AMT_log"]][validation_index])

rmse<-sqrt(mean((real_value-prediction)^2))

correlation<-cor(prediction,real_value)

plot(prediction,real_value)
abline(0,1, col=2)

Stack

stack_model<-create_stack_model_mlr(df_encoded,"DONATION_AMT_log",test_index,mlr_list)
result<-stack_test_mlr(stack_model,mlr_list,df_encoded,validation_index,"DONATION_AMT_log","log")

Check Stack Result (Correlation, plot)

prediction_value<-result$prediction_value
true_value<-result$true_value

correlation<-cor(prediction_value,true_value)

plot(prediction_value,true_value)
abline(0,1, col=2)