207 

 

 

THE EFFECTIVENESS ANALYSIS OF RANDOM FOREST ALGORITHMS 
WITH SMOTE TECHNIQUE IN PREDICTING LUNG CANCER RISK 

 
Ita Yulianti1, Ami Rahmawati2*), Tati Mardiana3 

 
Sistem Informasi Akuntansi Kampus Kota Sukabumi 

Universitas Bina Sarana Informatika 
ita.iyi@bsi.ac.id1 

 
Sistem Informasi2*), Sains Data3 

Universitas Nusa Mandiri 
ami.amv@nusamandiri.ac.id2*), tati.ttm@nusamandiri.ac.id3 

 
 (*) Corresponding Author 

 
Abstrak 

Jika dibandingkan dengan jenis kanker lainnya, sebagian besar penduduk penderita kanker meninggal karena 
kanker paru-paru. Seseorang perlu melakukan tes skrining melalui rontgen, CT scan, dan MRI untuk 
mendeteksi penyakitnya. Namun, sebelum melakukan proses tersebut, dokter biasanya akan melakukan 
anamnesis dan pemeriksaan fisik terlebih dahulu untuk mempelajari gejala dan kemungkinan faktor risiko 
kanker paru-paru. Kumpulan data kanker paru memiliki ketidakseimbangan kelas sehingga mempengaruhi 
kinerja algoritma  random forest dalam memprediksi risiko kanker paru.  Penelitian ini bertujuan untuk 
menerapkan  teknik SMOTE untuk meningkatkan kinerja algoritma random forest dalam memprediksi risiko 
kanker paru. Pada penelitian ini pengolahan dan analisis data  menggunakan bahasa pemrograman Python. 
Hasil pengujian menunjukkan nilai akurasi sebesar 88% dengan nilai AUC sebesar 0,93. Saat menggunakan 
algoritma random forest untuk memperkirakan risiko kanker paru-paru, teknik SMOTE berguna dalam 
menangani ketidakseimbangan kelas dalam kumpulan data. 
 
Kata kunci: Kanker Paru-Paru, Python, Random Forest, SMOTE 
 

Abstract 
When compared with other types of cancer, most of the population with cancer die from lung cancer.A person 
needs to do a screening test through X-rays, CT scans, and MRI to detect the disease. However, before carrying 
out the process, the doctor will ordinarily investigate a medical history and physical examination first to study 
the symptoms and possible risk factors for lung cancer. The lung cancer data set has a class imbalance that 
affects the performance of the random forest algorithm in predicting the risk of lung cancer. This study aims 
to employ the SMOTE technique to the random forest algorithm to increase accuracy in predicting lung cancer 
risk. In this research, data processing and analysis use the Python programming language. The test results 
show an accuracy value of 88% with an AUC value of 0.93. When employing the random forest method to 
forecast lung cancer risk, the SMOTE technique is useful in dealing with class imbalances in the data set. 
 
Keywords: Lung Cancer, Python, Random Forest, SMOTE 
 
 

INTRODUCTION 
 
Cancer is one of the leading causes of 

significant morbidity and mortality worldwide 
(Sofia & Tahlil, 2018). In 2018, the number of deaths 
from cancer reached 9.6 million people. Lung, 
prostate, colorectal, stomach, liver, and breast 
cancers are the biggest causes of cancer deaths 
every year (WHO, 2022). When compared with 
other types of cancer, most of the population with 
cancer died from lung cancer (Rattan et al., 2018). 

According to WHO data from 2020, lung cancer is 
the leading cause of death, with 1.80 million cases 
(WHO, 2021).  

In Indonesia itself, the majority of lung 
cancer sufferers are male due to smoking habits 
which can make the risk of getting cancer higher 
(Bulan et al., 2017). Lung cancer is difficult to detect 
because many cases of this cancer appear and show 
symptoms when its development has reached a 
certain stage (Makaju et al., 2018). Screening tests 
through X-rays, CT scans, and MRIs can detect the 



 

 
208 

 

 

disease in the lungs (Kurnia et al., 2016). However, 
before carrying out this process, the doctor will 
usually do a medical history and physical 
examination first to study the symptoms and 
possible risk factors for lung cancer (American 
Cancer Society, 2022).   

In line with the development of science and 
technology today, various fields ranging from 
education, economics, social, health, and many 
others are increasingly taking advantage of the role 
of technology, especially in the decision-making 
process. The many applications of this technology 
certainly encourage researchers to facilitate human 
work, for example in the health sector, namely to 
predict certain diseases using data mining 
techniques. This technique has many 
methods/algorithms that can be used, one of which 
is a random forest which is a modified algorithm 
from the decision tree. This approach has the 
advantage of being able to process a random 
selection of features in image datasets or in the form 
of attributes/parameters, resulting in a low error 
rate (Sari et al., 2020). 

The random forest algorithm is generally 
widely used in previous studies including to 
measure the severity of disease in apple leaf images 
(Ratnawati & Sulistyaningrum, 2019). In addition, 
this algorithm is also better than the xgboost 
algorithm in overcoming the imbalance class in the 
classification of hepatitis C disease levels (Syukron 
et al., 2020). Therefore, the purpose of this study 
was to analyze and measure the effectiveness of the 
random forest algorithm in predicting the risk of 
lung cancer through a physical examination in the 
form of perceived symptoms and habits or lifestyles. 
Patients with lung cancer usually have symptoms 
including shortness of breath, chest pain that does 
not improve, cough with blood, changes in the shape 
of the fingers, difficulty swallowing, and weight loss 
(Syifa et al., 2016). Not only that but in this study, the 
RF algorithm will also be integrated with the SMOTE 
technique to overcome if there is a class imbalance 
in the dataset used so that it can improve the 
performance of the algorithm and the predicted 
outcome obtained is optimal. 

This research is expected to provide 
novelty in the form of modeling that can be 
developed into a system that can predict the disease. 
Although not all lung cancer can be prevented, 
knowing the early detection of the risk of this 
disease, allows a person to immediately carry out a 
follow-up examination (screening) and be treated 
by a doctor so that it can reduce the risk of lung 
cancer. 

 
 

RESEARCH METHODS 

 
The following is an overview of the research 

framework used to describe all the stages carried 
out in this research: 

 
Figure 1. Research Methode framework 

 
The figure above describes the stages or 

steps taken to achieve the results of this research. 
This stage begins with selecting a dataset according 
to the topic raised, namely lung cancer. The next 
stage is preprocessing which includes removing 
duplicates, feature scaling and over-sampling smote. 
In the preprocessing stage, the dataset is processed 
first before entering the modeling stage. This is done 
so that the dataset used does not have problems that 
can complicate the processing. Then move on to the 
modeling stage, the dataset is divided into training 
data and testing data with a ratio of 80:20 using split 
validation combined with cross validation using the 
tested algorithm, namely random forest. Finally, the 
results of the algorithm test are validated with the 
help of a confusion matrix which gives output in the 
form of accuracy values and AUC graphs. 

 
Data Collection 

The dataset in this research is a secondary 
dataset obtained from the Kaggle Repository with 
CSV format which has a total of 16 attributes 
including the target class. 

 



 

 
209 

 

 

Preprocessing 
In this research, there are three stages of 

preprocessing the data used, namely, removing 
duplicates, feature scaling and over-sampling smote. 
The purpose of removing duplicates is to prevent 
duplication of data by filtering and deleting the same 
data in the dataset (Hendra & Fitriyani, 2021). 
Meanwhile, feature scaling is carried out to examine 
the diversity of values that occur in each variable 
and to balance the scale so that it has the same range 
of values (Aripin, 2021). 

Class balance in the application of 
classification algorithms is important to pay 
attention to so that the resulting performance has a 
good prediction. Therefore, to overcome the class 
imbalance in the research dataset, a resampling 
technique is used, namely oversampling, this 
technique was chosen because it can balance the 
dataset that is lacking in the minority class without 
reducing the dataset (Sulistiyono et al., 2021). The 
oversampling algorithm used is Synthetic Minority 
Over-sampling Technique (SMOTE), where this 
algorithm is an algorithm that can be called similar 
to random oversampling (ROS) but the difference is 
that in SMOTE, existing samples are not duplicated 
randomly but are made with the nearest neighbor 
concept (Arifiyanti & Wahyuni, 2020). The workings 
of the SMOTE algorithm are to interpolate the 
original data and then enter the artificial data 
generated in the minority class so that the data 
obtained varies (Indrawati, 2021). In general, the 
function of this algorithm is written with SMOTE (X, 
N, k), where X is the minority data, while N is the 
percentage of the total instances to be created, and k 
is the number of closest instances of the instance 
being searched for using the Euclidean distance 
formula as following (Sulistiyono et al., 2021):  
𝑑𝑖𝑠𝑡 =

 √(𝑋1 − 𝑌1)
2 + (𝑋2 − 𝑌2)

2 + ⋯ + (𝑋𝑛 − 𝑌𝑛 )
2…….(1) 

 
Modeling 

Random Forest was first introduced by 
Breiman in 2021, which is one of the decision tree-
based machine learning methods that is often used 
because it has the advantage of having high 
dimensions with a faster process that functions, 
especially on subset features (Ardiningtyas & 
Paulina Heruningsih Prima, 2021). This algorithm 
can be explained as a non-parametric model with 
the concept of combining learning models to 
improve performance in both regression and 
classification cases (Aripin, 2021). 

There are three steps in the development of 
the random forest algorithm, namely (Religia et al., 
2021): 
1. Set sampling is part of training k, 

2. Making each decision tree model, and  
3. Collection of k trees into the RF Model 

 
The random forest algorithm is a 

development method from the CART decision tree, 
so it is not surprising that in building the decision 
tree the CART method is used which begins by 
finding the entropy value first (Aripin, 2021) using 
Equation (2) below: 
𝐸𝑛𝑡𝑟𝑜𝑝𝑦 (𝑌) =  − ∑ 𝑖𝑝(𝑌)𝑙𝑜𝑔2𝑝(𝑌)  ......................... (2) 
 

Then proceed with calculating the 
information gain using Eq. (3) 
𝐼𝑛𝑓𝑜𝑟𝑚𝑎𝑡𝑖𝑜𝑛 𝐺𝑎𝑖𝑛 (𝑌, 𝑎) =  𝐸𝑛𝑡𝑟𝑜𝑝𝑦(𝑌) −

∑ 𝑣𝑎𝑙𝑢𝑒𝑠(𝑎)
|𝑌𝑣|

|𝑌𝑎|
(𝐸𝑛𝑡𝑟𝑜𝑝𝑦(𝑌𝑣))  ................................  (3) 

 
 

RESULTS AND DISCUSSION 
 

Dataset 
Collecting data that is processed in this 

study is public data obtained from the Kaggle 
website. The dataset was collected from the Lung 
Cancer Prediction System's online site. The 
following is a description of the lung cancer dataset 
as shown in Table 1. 
 

Table 1. Lung Cancer Dataset Description 
Lung Cancer 

Dataset 
Number of 
Attributes 

Number of 
Instances 

16 309 
 
From the data obtained, there are attributes, 
namely: gender, age, smoking, yellow fingers, 
anxiety, peer pressure, chronic disease, fatigue, 
allergy, wheezing, alcohol consuming, coughing, 
shortness of breath, swallowing difficulty, chest 
pain, and lung cancer. An explanation of each 
attribute will be described in Table 2. 
 

Table 2. Description of Data Attributes 

Attributes Value 

Gender M (Male), F (Female) 

Age 21 – 87 

Smoking Yes = 2, No = 1 

Yellow Fingers Yes = 2, No = 1 

Anxiety Yes = 2, No = 1 

Peer Pressure Yes = 2, No = 1 

Chronic Disease Yes = 2, No = 1 

Fatigue Yes = 2, No = 1 

Allergy Yes = 2, No = 1 

Wheezing Yes = 2, No = 1 

Alcohol Yes = 2, No = 1 



 

 
210 

 

 

Coughing Yes = 2, No = 1 

Shortness Of Breath Yes = 2, No = 1 

Swallowing Difficulty Yes = 2, No = 1 

Chest Pain Yes = 2, No = 1 

Class Lung Cancer Yes, No 

 
Data Exploration and Visualization 

Data exploration and visualization are 
carried out to obtain information and understand 
the attributes in the dataset. Exploration and 
visualization of several attributes are presented in 
graphical form as follows: 

 
Figure 2. Age Attribute 

 
The results in the first graph (See Figure 2.) show 
that the age attribute that dominates people at risk 
for lung cancer is the age of 60-70 years. 

 

 
Figure 3. Yellow Fingers Attribute 

 

Then, the second result is in the form of a pie chart 
that shows the percentage of yellowing fingers 
symptoms experienced by people who are at risk of 
lung cancer by 42.39%. 
 

 
Figure 4. Shortness of Breath Attribute 

 
Based on the results of exploration and visualization 
of further data, it shows that shortness of breath is a 
symptom that is often experienced by people who 
are at risk of lung cancer. 
 

 
Figure 5. Smoking Attribute 

 
Exploration of data for smoking attributes shows 
that active smokers are more susceptible to lung 
cancer. 
 



 

 
211 

 

 

 
Figure 6. Age dan Alcohol Attribute 

 
The next graph (See Figure 6.) proves that the age 
range of 50-80 years and consuming alcohol is more 
at risk of lung cancer. 
 

 
Figure 7. Coughing Attribute 

 
In addition, there is also another attribute that is the 
most common symptom experienced by people at 
risk for lung cancer, namely cough (See results of 
data exploration and visualization in Figure 7.). 
 

 
Figure 8. Wheezing Attribute 

 
Finally, the graphic results from the wheezing 
attribute show that loud, high-frequency breathing 
sounds are often experienced by people who are at 
risk for lung cancer. 
 
Modeling Results 
a. Remove Duplicate 

Removing duplicates is a step taken to 
remove duplication of data. In the lung cancer 
dataset, there are 33 duplicates, so removing 
duplicates is done. The number of tuples in the 
dataset, which was originally 309, has changed to 
276. 

 
b. Feature Scaling 

In this research, feature scaling is carried 
out, namely to make numerical data have the same 
range of values. The scaling feature used is 
MinMaxScaler to scale data values into a range. 

 
c. Synthetic Minority Over-sampling Technique 

(SMOTE) 

 
Figure 9. Class Lung Cancer 

Based on Figure 9, it can be seen that there 
is a very large distribution of data between classes, 
where the Yes class has a sample size of more than 



 

 
212 

 

 

250 data while the No class only has a sample 
number of fewer than 50 data so that the class is not 
balanced. To deal with imbalanced classes, in this 
study, SMOTE was used. SMOTE will select a point 
from the minority class and calculate the K nearest 
neighbors for this point. 
 
d. Modeling Results with Random Forest 

To determine and test the classification 
model, a data split was carried out, namely dividing 
the data into two parts. 80% for training data and 
20% for test data. The training data is used to build 
the model while the test data is used for valid 
performance evaluation. In addition, cross-
validation was also carried out on the test data with 
a value of K = 10. The results of the tests carried out 
were to produce values of accuracy (confusion 
matrix), precision, recall, and AUC. The results of the 
confusion matrix can be seen in Table 3. 
 

Table 3. Confusion Matrix of RF Algorithm 

Classification 
Classified as 

Positive Negative 
Positive 4 5 
Negative 2 45 

 
Based on Table 3, there are details on the number of 
True Positive = 4, False Positive = 5, False Negative 
= 2, and True Negative = 45. From this data, 
accuracy, precision, and recall can be calculated. The 
test result data can be seen in Figure 10 and Table 4. 
 

Table 4. Results of Random Forest Test  
Performance Random Forest 

Accuracy 0,88 
Precision 0,90 
Recall 0,95 

 

 
Figure 10. Results of Random Forest Test 

 
In addition, the test results for the random forest 
algorithm can also be seen based on the ROC graph 
which can be seen through the resulting AUC value 
of 0.93. This proves that the accuracy of the model is 
included in the excellent classification category. The 
resulting ROC graph can be seen in Figure 11. 
 

 
Figure 11. AUC Random Forest 

 
Meanwhile, the precision and recall graphs are 
visualized through a graph that can be seen in the 
following figure. 
 

 
Figure 12. PRC Random Forest 

 
 

CONCLUSIONS AND SUGGESTIONS 
 
Conclusion 

From the research results obtained, the 
lung cancer dataset has a class imbalance. To 
overcome the class imbalance, SMOTE (Minority 
Over Sampling Technique) was used and the 
random forest algorithm was used for classification 
testing. Testing the lung cancer dataset yielded an 
accuracy of 88% with an AUC value of 0.93. While 
the precision value is 0.90 and the recall value is 
0.95. Based on this explanation, it can be concluded 
that the application of the SMOTE technique to the 
random forest algorithm can improve the 
performance of unbalanced data classification, so 
that the level of effectiveness of the resulting 
algorithm test becomes more optimal. 
 
Suggestion 



 

 
213 

 

 

For further research, it is hoped that the 
modeling resulting from this research can be 
redeveloped by being integrated using optimization 
algorithms and implemented into a GUI application 
that can be easily accessed by everyone, as well as 
adding image datasets so that prediction results are 
better. 

 
 

REFERENCES 
 
American Cancer Society. (2022). Lung Cancer. 

American Cancer Society. 
https://www.cancer.org/cancer/lung-
cancer/ 

Ardiningtyas, Y. E., & Paulina Heruningsih Prima, R. 
(2021). ANALISIS BALANCING DATA UNTUK 
MENINGKATKAN AKURASI DALAM 
KLASIFIKASI. 24–28. 

Arifiyanti, A. A., & Wahyuni, E. D. (2020). SMOTE : 
METODE PENYEIMBANG KELAS PADA 
KLASIFIKASI DATA MINING. XV, 34–39. 

Aripin, H. A. (2021). Jurnal Informatika dan 
Komputer (INFOKOM). 9, 30–45. 

Bulan, I. A. K. T., Ratnawati, H., Wargasetia, T. L., 
Bulan, I. A. K. T., Ratnawati, H., & Wargasetia, T. 
L. (2017). Gambaran pasien kanker paru di 
rumah sakit Immanuel Bandung periode 
januari 2013 hingga desember 2014. Journal of 
Medicine and Health, 1(6), 517–524. 

Hendra, A., & Fitriyani. (2021). Analisis Sentimen 
Review Halodoc Menggunakan Nai ̈ ve Bayes 
Classifier. 6(2), 78–89. 

Indrawati, A. (2021). PENERAPAN TEKNIK 
KOMBINASI OVERSAMPLING DAN 
UNDERSAMPLING HYBRID OVERSAMPLING 
AND UNDERSAMPLING TECHNIQUES TO 
HANDLING IMBALANCED DATASET. 4(1), 38–
43. https://doi.org/10.33387/jiko 

Kurnia, R., Rahmadewi, R., & Aini, F. (2016). Deteksi 
Dini Penyakit Paru Dengan Metoda Bayesian 
Berbasis Android. National Conference of 
Applied Engineering, Business and Information 
Technology, Politeknik Negeri Padang, 317–
323. 

Makaju, S., Prasad, P. W. C., Alsadoon, A., Singh, A. K., 
& Elchouemi, A. (2018). Lung Cancer Detection 
using CT Scan Images. Procedia Computer 
Science, 125(2009), 107–114. 
https://doi.org/10.1016/j.procs.2017.12.016 

Ratnawati, L., & Sulistyaningrum, D. R. (2019). 

Penerapan Random Forest untuk Mengukur 
Tingkat Keparahan Penyakit. 8(2). 

Rattan, S., Kaur, S., Kansal, N., & Kaur, J. (2018). An 
optimized lung cancer classification system for 
computed tomography images. 2017 4th 
International Conference on Image Information 
Processing, ICIIP 2017, 2018-Janua, 15–20. 
https://doi.org/10.1109/ICIIP.2017.8313676 

Religia, Y., Nugroho, A., & Hadikristanto, W. (2021). 
JURNAL RESTI Analisis Perbandingan 
Algoritma Optimasi pada Random Forest untuk. 
1(10), 187–192. 

Sari, V. R., Firdausi, F., & Azhar, Y. (2020). 
Perbandingan Prediksi Kualitas Kopi Arabika 
dengan Menggunakan Algoritma SGD, Random 
Forest dan Naive Bayes. EDUMATIC : Jurnal 
Pendidikan Informatika, 4(2), 1–9. 
https://doi.org/10.29408/edumatic.v4i2.220
2 

Sofia, R., & Tahlil, T. (2018). Pengalaman Pasien 
Kanker dalam Menghadapi Kemoterapi. Jurnal 
Ilmu Keperawatan, 6(2), 81–91. 

Sulistiyono, M., Pristyanto, Y., Adi, S., & Gumelar, G. 
(2021). Implementasi Algoritma Synthetic 
Minority Over-Sampling Technique untuk 
Menangani Ketidakseimbangan Kelas pada 
Dataset Klasifikasi. SISTEMASI, 10(2), 445. 
https://doi.org/10.32520/stmsi.v10i2.1303 

Syifa, R. A., Adi, K., Fisika, D., & Diponegoro, U. 
(2016). Analisis Tekstur Citra Mikroskopis 
Kanker Paru Menggunakan Metode Gray Level 
Co-Occurance Matrix (Glcm) Dan Tranformasi 
Wavelet Dengan Klasifikasi Naive Bayes. 
Youngster Physics Journal, 5(4), 457–462. 

Syukron, M. ;, Santoso, R., & Widiharih, T. (2020). 
ISSN: ISSN: 2339-2541JURNAL GAUSSIAN, 
Volume9, Nomor 3, Tahun 2020, Halaman 
227-236Online di: 
https://ejournal3.undip.ac.id/index.php/gaus
sian/PERBANDINGAN METODE SMOTE 
RANDOM FOREST DAN SMOTE XGBOOST 
UNTUK KLASIFIKASI TINGKAT PENYAKIT 
HEPATITIS C PADA IM. Jurnal Gaussian, 9(3), 
227–236. 

WHO. (2021). Cancer. World Health Organization. 
https://www.who.int/news-room/fact-
sheets/detail/cancer 

WHO. (2022). Cancer. World Health Organization. 
https://www.who.int/health-
topics/cancer#tab=tab_1