김세중 노현지 양다인 오진영

The risk of traffic accident deaths in Korea is decreasing year by year.

However, Korea still has a high risk, ranking 27th in the number of traffic accident deaths compared to the population among 36 OECD countries.

Therefore, our team aimed to reduce the risk of traffic accidents by creating a model that predicts the risk of traffic accidents in certain areas of Seoul to help people go out at safer times and to strengthen crackdowns and guide them to be alert during high-risk times

114442 rows x 10 Columns

• 발생일
• 발생시간
• 발생지_시도
• 발생지_시군구
• 법정동명
• 사고건수
• 사망자수
• 중상자수
• 경상자수
• 부상신고자수

450 rows x 14 Columns

• 동별(1)
• 동별(2)
• 동별(3)
• 2017 1/4
• 2017 2/4
• 2017 3/4
• 2017 4/4
• 2018 1/4
• 2018 2/4
• 2018 3/4
• 2018 4/4
• 2019 1/4
• 2019 2/4
• 2019 3/4
• 2019 4/4

765 rows x 7 Columns

• 행정동코드
• 시도명
• 시군구명
• 읍면동명
• 법정동코드
• 동리명
• 생성일자

180 rows x 2

• population_regin
• mapping_region

• Cleaning dirty data
• Feature creation derived from '발생일'
• Derive New '인구' features by aggregatingg over two entities
• Featrue creation '위험도' columns in accident_df
• '법정동명', '발생_요일' with LabelEncoder

• Create a new feature related to multiple columns using k-means clustering.
• Perform outliner detection to turn the unbalanced distribution of categories A through D into a balanced one.

• Create a predictive model using RandomForest Algorithm.
• Use GridSearchCV to find the most appropriate hyperparameters.
• Using k-fold croos validation for testing. The average score was 57%.

• The model was trained based on the clustering of the target features.
• Create a predictive model using DesicionTree Classification.
• Using k-fold croos validation for testing. The average score was 65%.