SpaceX Falcon9 First Stage Landing Prediction

Kaggle Link : https://www.kaggle.com/code/yoontaeklee/spacex-falcon-9-firsts-stage-landing-prediction

import numpy as np
import pandas as pd
import missingno as msno
import matplotlib.pyplot as plt
import seaborn as sns

from sklearn import preprocessing

# Allows us to split out data into training and testing data
from sklearn.model_selection import train_test_split

# Allows us to test parameters of classification algorithms and find the best one
from sklearn.model_selection import GridSearchCV

# Logistic Regression classification algorithm
from sklearn.linear_model import LogisticRegression

# Support Vector Machine classification algorithm
from sklearn.svm import SVC

# Decision Tree classification algorithm
from sklearn.tree import DecisionTreeClassifier

# K Nearest Neighbors classification algorithm
from sklearn.neighbors import KNeighborsClassifier

Process

1. 데이터셋 확인 (null data)
2. EDA (Exploratory Data Analysis)
3. Feature Engineering (one-hot encoding, class 나누기, 구간 나누기 등)
4. Model 만들기

1. 데이터셋 확인 (null data)

df = pd.read_csv("https://cf-courses-data.s3.us.cloud-object-storage.appdomain.cloud/IBM-DS0321EN-SkillsNetwork/datasets/dataset_part_1.csv")
df.head(10)

df.describe()

1.1 Null data 확인

for column in df.columns:
		msg = 'column: {:>10}\t Percent of NaN Value: {:.2f}%'
				.format(column, 100 * (df[column].isnull().sum()) / df[column].shape[0])
		print(msg)

• missingno 라이브러리 통해 더 쉽게 null data 확인

msno.matrix(df=df.iloc[:,:], figsize=(7,7), color=(0.3,0.3,0.3))

msno.bar(df=df.iloc[:,:], figsize=(7,7), color=(0.3,0.3,0.3))

1.2 Target label 확인

• Outcome 의미
  • True Ocean : 특정 바다에 성공적으로 착륙
  • False Ocean : 특정 바다에 착륙 실패
  • True RTLS : 특정 Ground Pad에 성공적으로 착륙
  • False RTLS : 특정 Ground Pad에 착륙 실패
  • True ASDS : 드론쉽에 성공적으로 착륙
  • False ASDS : 드론쉽에 착륙 실패
  • None ASDS, None None : 착륙 실패

landing_outcomes = df.Outcome.value_counts()
print(landing_outcomes)

for i, outcome in enumerate(landing_outcomes.keys()):
		print(i,outcome)

• 이전의 타이타닉 프로젝트와 마찬가지로, 최종 결과만 중요 ⇒ 성공, 실패로만 분류
  

bad_outcomes = set(landing_outcomes.keys()[[1,3,5,6,7]])

landing_class = []
for outcome in (df['Outcome']):
		if outcome in bad_outcomes:
				landing_class.append(0)
		else:
				landing_class.append(1)

df['Class'] = landing_class
df[['Class']].head(10)

2. EDA (Exploratory Data Analysis)

• 모델에 쓸만한 feature 탐색

2.1 FlightNumber vs Payload

sns.catplot(y="PayloadMass",x="FlightNumber",hue="Class",data=df,aspect=5)
plt.xlabel("Flight Number", fontsize=22)
plt.ylabel("Pay load Mass (kg)", fontsize=22)
plt.show()

2.2 FlightNumber vs LaunchSite

sns.catplot(y="LaunchSite",x="FlightNumber",hue="Class",data=df,aspect=5)
plt.xlabel("Flight Number",fontsize=22)
plt.ylabel("Launch Site",fontsize=22)
plt.show()

2.3 Payload vs LaunchSite

sns.catplot(y="LaunchSite",x="PayloadMass",hue="Class",data=df,aspect=5)
plt.xlabel("Payload Mass (KG)", fontsize=22)
plt.ylabel("Launch Site",fontsize=22)
plt.show()

• 탑재량이 많은 경우 CCAFS SLC 40 혹은 KSC LC 39A 에서 발사함
• CCAFS SLC 40 에서 발사하는 로켓 중, 8000kg 미만을 탑재하는 경우 더 높은 실패율을 보임

2.4 Orbit에 따른 성공률

df.groupby("Orbit").mean()["Class"].plot(kind='bar')
plt.xlabel("Orbit Type",fontsize=20)
plt.ylabel("Success Rate",fontsize=20)
plt.show()
![](https://velog.velcdn.com/images/hsjunior1/post/0d70e847-5fa5-44b2-85f7-9d05ac3f74cd/image.png)

2.5 FlightNumber vs Orbit Type

sns.catplot(y="Orbit",x="FlightNumber",hue="Class",data=df,aspect=5)
plt.xlabel("Flight Number",fontsize=20)
plt.ylabel("Orbit Type",fontsize=20)
plt.show()
![](https://velog.velcdn.com/images/hsjunior1/post/ca320a63-f5f5-49e3-bc9e-98e3399cb33c/image.png)

• LEO의 경우 횟수와 성공이 비례
• GTO의 경우 횟수와 성공이 무관

2.6 Orbit Type vs Payload

sns.catplot(y="Orbit",x="PayloadMass",hue="Class",data=df,aspect=5)
plt.ylabel("Orbit type",fontsize=20)
plt.xlabel("Pay load (Kg)")
plt.show()
![](https://velog.velcdn.com/images/hsjunior1/post/e1b154d1-c09f-4e8a-9221-afd004c137c5/image.png)

• GTO 궤도의 경우, 적재량이 많을수록 성공률이 떨어짐

2.7 년도와 성공률과의 관계

• 현재 갖고있는 날짜 데이터에서 년도에 해당하는 부분만 추출

years = []
def Extract_year(date):
		for i in df["Date"]:
				years.append(i,split("-")[0]
		return years

df1 = pd.DataFrame(Extract_year(df["Date"]), columns = ["year"])
df1["Class"] = df["Class"]
sns.lineplot(y=df1.groupby('year')['Class'].mean(),
		x=np.unique(Extract_year(df['Date'])))
plt.xlabel("Year", fontsize = 20)
plt.ylabel("Success Rate", fontsize = 20)
plt.show()

3. Feature Engineering

• 각 feature가 성공률에 얼마나 큰 영향을 미치는지 확인하고, 후에 모델에 사용

features = df[['FlightNumber','PayloadMass','Orbit','LaunchSite','Flights','GridFins','Reused','Legs','LandingPad','Block','ReusedCount','Serial']]
features.head()

3.1 One hot encoding

features_one_hot = pd.get_dummies(features, columns = ['Orbit','LaunchSite','LandingPad','Serial'])
features_one_hot.head()

features_one_hot.astype('float64')

4. Building Machine Learning Model and prediction using the trained model

• 기본 df에 있는 ‘Class’ feature’에서 새로운 numpy array를 만든다

Y = df['Class'].to_numpy()
Y

transform = preprocessing.StandardScaler()
X = transform.fit_transform(features_one_hot)

• train_test_split 이용하여 트레이닝 데이터와 테스트 데이터 나누기

X_train, X_test, Y_train, Y_test = train_test_split(X,Y,test_size=0.2, random_state=2)
print('Train set: ',X_train.shape, Y_train.shape)
print('Test set: ',X_test.shape, Y_test.shape)

• GridSearchCV 로 파라미터 최적화

parameters = {'C':[0.01,0.1,1], 'penalty':['l2'], 'solver':['lbfgs']}
lr = LogisticRegression()

logreg_cv = GridSearchCV(lr,parameters,cv=10)
logreg_cv.fit(X_train,Y_train)

print('best parameters : ',logreg_cv.best_params_)
print('accuracy : ',logreg_cv.best_score_)

print('test set accuracy : ',logreg_cv.score(X_test, Y_test))

yhat = logreg_cv.predict(X_test)
plot_confusion_matrix(Y_test,yhat)

• 여기서 중요한 것은, 성공 실패가 아님 ⇒ 성공할 것이라고 예측했으나 실패한 경우가 주목해야 하는 부분
  

parameters = {'kernel':('linear','rbf','poly','rbf','sigmoid'),'C':np.logspace(-3,3,5),'gamma':np.logspace(-3,3,5)}
svm = SVC()

svm_cv = GridSearchCV(svm, parameters, cv=10)
svm_cv.fit(X_train, Y_train)

print("best parameters : ",svm_cv.best_params_)
print("accuracy : ",svm_cv.best_score_)

print("test set accuracy : ",svm_cv.score(X_test, Y_test))

yhat = svm_cv.predict(X_test)
plot_confusion_matrix(Y_test,yhat)

parameters = {'criterion': ['gini','entropy'],
             'splitter':['best','random'],
             'max_depth':[2 * n for n in range(1,10)],
             'max_features':['auto','sqrt'],
             'min_samples_leaf':[1,2,4],
             'min_samples_split':[2,5,10]}
tree = DecisionTreeClassifier()

tree_cv = GridSearchCV(tree, parameters, cv = 10)
tree_cv.fit(X_train, Y_train)

print("best parameters : ",tree_cv.best_params_)
print("accuracy : ",tree_cv.best_score_)

print("test set accuracy : ",tree_cv.score(X_test,Y_test))

yhat = tree_cv.predict(X_test)
plot_confusion_matrix(Y_test,yhat)

parameters = {'n_neighbors': [1, 2, 3, 4, 5, 6, 7, 8, 9, 10],
              'algorithm': ['auto', 'ball_tree', 'kd_tree', 'brute'],
              'p': [1,2]}
KNN = KNeighborsClassifier()

knn_cv = GridSearchCV(KNN,parameters,cv=10)
knn_cv.fit(X_train, Y_train)

print("best parameters : ",knn_cv.best_params_)
print("accuracy : ",knn_cv.best_score_)

print("test set accuracy : ",knn_cv.score(X_test,Y_test))

yhat = knn_cv.predict(X_test)
plot_confusion_matrix(Y_test,yhat)

• 모든 테스트 결과, Decision Tree Classification 의 방법이 가장 높은 정확도를 보였다