Scribd
Upload
4 views13 pages

XGBoost and CNN for Engine Status Prediction

The document details the implementation of various machine learning models, including XGBoost, CNN, Random Forest, and SVM, for predicting engine status based on multiple parameters. It involves data preprocessing steps such as handling missing values, standardizing features, and applying SMOTE for class imbalance. The models are evaluated using accuracy and confusion matrices, with XGBoost achieving the highest accuracy of 0.9744, followed by CNN at 0.9593, Random Forest at 0.9576, and SVM at 0.8616.

Uploaded by

vihar joshi
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
4 views13 pages

XGBoost and CNN for Engine Status Prediction

The document details the implementation of various machine learning models, including XGBoost, CNN, Random Forest, and SVM, for predicting engine status based on multiple parameters. It involves data preprocessing steps such as handling missing values, standardizing features, and applying SMOTE for class imbalance. The models are evaluated using accuracy and confusion matrices, with XGBoost achieving the highest accuracy of 0.9744, followed by CNN at 0.9593, Random Forest at 0.9576, and SVM at 0.8616.

Uploaded by

vihar joshi
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd

# XGBoost

import pandas as pd
import numpy as np
import seaborn as sns
import [Link] as plt
from sklearn.model_selection import train_test_split
from [Link] import StandardScaler
from [Link] import accuracy_score, confusion_matrix,
classification_report
from collections import Counter
import xgboost as xgb
from imblearn.over_sampling import SMOTE

# Load dataset
file_path = "c:/Users/TATA User/Downloads/HTB Data (1).xlsx"
df = pd.read_excel(file_path, sheet_name="Data")

# Select main parameters only


main_params = [
'P_Luboil', # Oil pressure higher
'Eng SFC', # SFC
'Lambda', # Lambda
'EngPower', # Power
'E_Torque', # Torque
'BlowBy', # Blow by pressure
'P_Cr_Case', # Crank case pressure
'T_LubOil', # Oil temperature idle
'T_Heater', # Oil temperature higher
'Status' # Target variable
]
df = df[main_params]

# Clean and convert target variable to binary (OK = 1, Nok = 0)


df['Status'] = df['Status'].astype(str).[Link]().replace({'OK': 1,
'Ok': 1, 'Nok': 0, 'NOK': 0})

# Check unique values and distribution


print("Unique values in 'Status':", df['Status'].unique())
print("Class distribution:\n", df['Status'].value_counts())
# Handle missing values (fill with mean for numeric columns)
for col in [Link]:
if col != 'Status':
df[col] = df[col].fillna(df[col].mean())

# Ensure no NaN values in Status


df['Status'] = df['Status'].fillna(df['Status'].mode()[0])

# Define features and target


X = [Link](columns=['Status'])
y = df['Status'].astype(int) # Ensure target is integer

# Standardize numerical features


scaler = StandardScaler()
X = scaler.fit_transform(X)

# Apply SMOTE to handle class imbalance


smote = SMOTE(random_state=42)
X_resampled, y_resampled = smote.fit_resample(X, y)

# Split dataset
X_train, X_test, y_train, y_test = train_test_split(X_resampled,
y_resampled, test_size=0.2, random_state=42, stratify=y_resampled)

# Define and train optimized XGBoost model


xgb_model = [Link](
objective="binary:logistic",
eval_metric="logloss",
use_label_encoder=False,
n_estimators=300, # Increased trees for better learning
learning_rate=0.03, # Slower learning for better generalization
max_depth=8, # Increased depth for capturing complex
patterns
min_child_weight=3, # Prevents overfitting on noisy data
gamma=0.1, # Helps in pruning unnecessary splits
subsample=0.85, # Helps with generalization
colsample_bytree=0.9, # Uses more features per tree
scale_pos_weight=len(y_train[y_train == 0]) / len(y_train[y_train
== 1]), # Handles class imbalance
random_state=42
)
xgb_model.fit(X_train, y_train)

# Predict probabilities for threshold tuning


y_pred_probs = xgb_model.predict_proba(X_test)[:, 1]

# **Optimized threshold tuning**


optimal_threshold = 0.6 # Adjust threshold to reduce FP & FN
y_pred_xgb = (y_pred_probs > optimal_threshold).astype(int)

# Evaluate the model


acc_xgb = accuracy_score(y_test, y_pred_xgb)
print(f"XGBoost Model - Accuracy: {acc_xgb:.4f}")
print(classification_report(y_test, y_pred_xgb, zero_division=1))

# **Improved Confusion Matrix Visualization**


# Compute confusion matrix
cm = confusion_matrix(y_test, y_pred_xgb)
total_predictions = [Link](cm) # Total number of samples

[Link](figsize=(6,5))
[Link](cm, annot=True, fmt='d', cmap='Blues',
xticklabels=['Engine Not OK (0)', 'Engine OK (1)'],
yticklabels=['Engine Not OK (0)', 'Engine OK (1)'])

# Add total predictions in the title


[Link]('Predicted')
[Link]('Actual')
[Link](f'Optimized Confusion Matrix - XGBoost Model\nTotal
Predictions: {total_predictions}')
[Link]()

XGBoost Model - Accuracy: 0.9744


precision recall f1-score support

0 0.97 0.98 0.97 430


1 0.98 0.97 0.97 429

accuracy 0.97 859


macro avg 0.97 0.97 0.97 859
weighted avg 0.97 0.97 0.97 859
CNN

# CNN

import pandas as pd
import numpy as np
import seaborn as sns
import [Link] as plt
from sklearn.model_selection import train_test_split
from [Link] import StandardScaler
from [Link] import accuracy_score, confusion_matrix,
classification_report
from collections import Counter
import tensorflow as tf
from tensorflow import keras
from [Link] import Sequential
from [Link] import Dense, Dropout, BatchNormalization
from imblearn.over_sampling import SMOTE

# Load dataset
file_path = "c:/Users/TATA User/Downloads/HTB Data (1).xlsx"
df = pd.read_excel(file_path, sheet_name="Data")

# Select main parameters only


main_params = [
'P_Luboil', # Oil pressure higher
'Eng SFC', # SFC
'Lambda', # Lambda
'EngPower', # Power
'E_Torque', # Torque
'BlowBy', # Blow by pressure
'P_Cr_Case', # Crank case pressure
'T_LubOil', # Oil temperature idle
'T_Heater', # Oil temperature higher
'Status' # Target variable
]
df = df[main_params]

# Clean and convert target variable to binary (OK = 1, Nok = 0)


df['Status'] = df['Status'].astype(str).[Link]().replace({'OK': 1,
'Ok': 1, 'Nok': 0, 'NOK': 0})

# Check unique values and distribution


print("Unique values in 'Status':", df['Status'].unique())
print("Class distribution:\n", df['Status'].value_counts())

# Handle missing values (fill with mean for numeric)


for col in [Link]:
if col != 'Status':
df[col] = df[col].fillna(df[col].mean())

# Ensure no NaN values in Status


df['Status'] = df['Status'].fillna(df['Status'].mode()[0])

# Define features and target


X = [Link](columns=['Status'])
y = df['Status'].astype(int) # Ensure target is integer

# Standardize numerical features


scaler = StandardScaler()
X = scaler.fit_transform(X)

# Apply SMOTE to handle class imbalance


smote = SMOTE(random_state=42)
X_resampled, y_resampled = smote.fit_resample(X, y)

# Ensure dataset contains both classes


if len([Link]()) < 2:
raise ValueError("Error: The dataset must contain both 'OK' and
'Nok' samples!")

# Split dataset
X_train, X_test, y_train, y_test = train_test_split(X_resampled,
y_resampled, test_size=0.2, random_state=42, stratify=y_resampled)

# Define improved CNN model


def build_cnn_model(input_shape):
model = Sequential([
Dense(128, activation='relu', input_shape=(input_shape,)),
BatchNormalization(),
Dropout(0.3),
Dense(64, activation='relu'),
BatchNormalization(),
Dropout(0.3),
Dense(32, activation='relu'),
BatchNormalization(),
Dropout(0.3),
Dense(1, activation='sigmoid')
])
[Link](optimizer=[Link](learning_rate=0.001),
loss='binary_crossentropy', metrics=['accuracy'])
return model

# Train and evaluate CNN model


cnn_model = build_cnn_model(X_train.shape[1])
early_stopping = [Link](monitor='val_loss',
patience=5, restore_best_weights=True)
cnn_model.fit(X_train, y_train, epochs=100, batch_size=32,
validation_data=(X_test, y_test), callbacks=[early_stopping],
verbose=1)

# Evaluate CNN model


y_pred_cnn = (cnn_model.predict(X_test) > 0.5).astype(int)
acc_cnn = accuracy_score(y_test, y_pred_cnn)
print(f"CNN Model - Accuracy: {acc_cnn:.4f}")
print(classification_report(y_test, y_pred_cnn, zero_division=1))

# Plot Confusion Matrix

cm = confusion_matrix(y_test, y_pred_xgb)
total_predictions = [Link](cm) # Total number of samples

[Link](figsize=(6,5))
[Link](cm, annot=True, fmt='d', cmap='Blues',
xticklabels=['Engine Not OK (0)', 'Engine OK (1)'],
yticklabels=['Engine Not OK (0)', 'Engine OK (1)'])

# Add total predictions in the title


[Link]('Predicted')
[Link]('Actual')
[Link](f'Optimized Confusion Matrix - XGBoost Model\nTotal
Predictions: {total_predictions}')
[Link]()
CNN Model - Accuracy: 0.9593
precision recall f1-score support

0 0.94 0.99 0.96 430


1 0.99 0.93 0.96 429

accuracy 0.96 859


macro avg 0.96 0.96 0.96 859
weighted avg 0.96 0.96 0.96 859
Random Forestclassifier and SVM

#RandomForestClassifier and SVM

import pandas as pd
import numpy as np
import seaborn as sns
import [Link] as plt
from sklearn.model_selection import train_test_split, cross_val_score
from [Link] import StandardScaler
from [Link] import RandomForestClassifier
from [Link] import SVC
from [Link] import accuracy_score, confusion_matrix,
classification_report
from collections import Counter

# Load dataset
file_path = "c:/Users/TATA User/Downloads/HTB Data (1).xlsx"
df = pd.read_excel(file_path, sheet_name="Data")

# Select main parameters only


main_params = [
'P_Luboil', # Oil pressure higher
'Eng SFC', # SFC
'Lambda', # Lambda
'EngPower', # Power
'E_Torque', # Torque
'BlowBy', # Blow by pressure
'P_Cr_Case', # Crank case pressure
'T_LubOil', # Oil temperature idle
'T_Heater', # Oil temperature higher
'Status' # Target variable
]
df = df[main_params]

# Convert target variable to binary (OK = 1, NOK = 0)


df['Status'] = df['Status'].astype(str).[Link]().map({'OK': 1,
'Nok': 0})

# Fix: Ensure at least one sample from each class exists in the dataset
if df['Status'].nunique() < 2:
raise ValueError("Error: The dataset must contain both 'OK' and
'NOK' samples!")

# Fix: Handle missing values correctly


df['Status'] = df['Status'].fillna(df['Status'].mode()[0]) # Fill NaN
in target column

# Fill missing values in features with mean


for col in [Link]:
if col != 'Status':
df[col] = df[col].fillna(df[col].mean())

# Define features and target


X = [Link](columns=['Status'])
y = df['Status']

# Standardize numerical features


scaler = StandardScaler()
X = scaler.fit_transform(X)

# Fix: Ensure stratified split has both classes


try:
X_train, X_test, y_train, y_test = train_test_split(
X, y, test_size=0.2, random_state=42, stratify=y
)
except ValueError as e:
print("Error in stratified split:", e)
print("Falling back to random split (not recommended for imbalanced
data).")
X_train, X_test, y_train, y_test = train_test_split(X, y,
test_size=0.2, random_state=42)

# Compute class weights for handling imbalance


class_counts = Counter(y_train)
total_samples = sum(class_counts.values())
class_weights = {cls: total_samples / (len(class_counts) * count) for
cls, count in class_counts.items()}

# Train and Evaluate Models


def evaluate_model(model, name):
[Link](X_train, y_train)
y_pred = [Link](X_test)
acc = accuracy_score(y_test, y_pred)
f1 = cross_val_score(model, X_train, y_train, cv=5,
scoring='f1_macro').mean()

print(f"{name} - Accuracy: {acc:.4f}, F1 Score: {f1:.4f}")


print(classification_report(y_test, y_pred, zero_division=1))

# Plot Confusion Matrix


cm = confusion_matrix(y_test, y_pred)
total_predictions = [Link](cm) # Total number of samples
[Link](cm, annot=True, fmt='d', cmap='Blues',
xticklabels=['Engine Not OK (0)', 'Engine OK (1)'],
yticklabels=['Engine Not OK (0)', 'Engine OK (1)'])
[Link]('Predicted')
[Link]('Actual')
[Link](f'Optimized Confusion Matrix - XGBoost Model\nTotal
Predictions: {total_predictions}')
[Link]()

# Models to test
models = {
"Random Forest": RandomForestClassifier(n_estimators=100,
random_state=42, class_weight=class_weights),
"SVM": SVC(kernel='rbf', probability=True,
class_weight=class_weights)
}

for name, model in [Link]():


evaluate_model(model, name)

Random Forest - Accuracy: 0.9576, F1 Score: 0.5011


precision recall f1-score support

0.0 1.00 0.00 0.00 19


1.0 0.96 1.00 0.98 429

accuracy 0.96 448


macro avg 0.98 0.50 0.49 448
weighted avg 0.96 0.96 0.94 448
SVM - Accuracy: 0.8616, F1 Score: 0.5598
precision recall f1-score support

0.0 0.16 0.53 0.24 19


1.0 0.98 0.88 0.92 429

accuracy 0.86 448


macro avg 0.57 0.70 0.58 448
weighted avg 0.94 0.86 0.89 448

You might also like