[ML] 집값 예측 인공지능 웹 API - 얼마요

https://my-apt.herokuapp.com/

• 랜덤포레스트 regressor를 사용한 예측 모델을 기반으로한 웹앱 서비스

introduction

This is Machine-learning Web API project for prediction of the value of apartments in Gwangmyung-si, Korea.

Environment & Skills

• MacOS
• Whale / Chrome Browser
• Python 3.8
  • Library for WebScraping : requests, APScheduler
  • Library for EDA & Data Preprocessing : Pandas, Numpy
  • Library for Data Visualization : Matplotlib, Plotly, Folium
  • Library for ML Models : sklearn, xgboost, lightgbm
• Flask, HTML, CSS
• Heroku
• Metabase
• ML MODELS
  • Tree-based-Models : RandomForest Regressor, Xgboost Regressor, Lightgbm Regressor
  • TransformedTargetRegressor : Log Transformed Linear Regression
• Miri Canvas
• OBS

pipeline

Data

광명시 공공 데이터 open api

• 광명시 아파트 실거래 공공데이터 request => MongoDB에 JSON형태로 적재 => SQLite3에 RDB형태로 변환 후 python api로 연동 => Pandas로 데이터프레임 형태로 불러오기

Evaluation

MODELS	MAE	RMSE	R2_SCORE
LinearRegression	174718437.51 KRW	238319972.99 KRW	0.08
ElasticNetCV	274784163.17 KRW	360266710.18 KRW	-1.1
RandomForestRegressor	63981882.48 KRW	113676925.82 KRW	0.9
XGB Regressor	149348580.28 KRW	180378806.28 KRW	0.47
LGBM Regressor	154223908.87 KRW	186687372.47 KRW	0.44