일 | 월 | 화 | 수 | 목 | 금 | 토 |
---|---|---|---|---|---|---|
1 | 2 | 3 | 4 | 5 | ||
6 | 7 | 8 | 9 | 10 | 11 | 12 |
13 | 14 | 15 | 16 | 17 | 18 | 19 |
20 | 21 | 22 | 23 | 24 | 25 | 26 |
27 | 28 | 29 | 30 |
- 딥러닝
- LG Aimers 4th
- 지도학습
- supervised learning
- 머신러닝
- LG Aimers
- 해커톤
- LG
- deep learning
- Classification
- GPT-4
- PCA
- OpenAI
- ChatGPT
- 분류
- Machine Learning
- 회귀
- regression
- 티스토리챌린지
- gpt
- 오블완
- LLM
- AI
- Today
- Total
SYDev
[캡스톤디자인] 시계열 예측 모델(LSTM, ARIMA, ..) 기초 본문
Time Series?
- Time Series: In mathematics, a time series is a series of data points indexed (or listed or graphed) in time order
- thus it is a sequence of discrete-time data
- 간격이 동일하지 않으면 Time Series Data X
Characteristics of Time Series Data
- A time series ccan be stationary or non-stationary
- Stationary Time Series: has statistical properties that are constant over time
- statistics like the mean, variance, autocorelation don change over the data
- most statistical methods, including ARIMA, are based on the assumption that the series can be mad approximately stationary through one or more transformations(한 번 이상의 변환을 통해서 data가 stationary를 만족할 수 있다고 가정)
- comparatively easy to predict
- non-stationary data is possible but difficult with an approach like ARIMA
- Another key feature of time series data is whether it has a trend present in the data
- ex) the prices in a grocery store from the last 50 years would exhibit a trend because inflation would drive those prices higher
- Predicting data that contains trends can be difficult
- if the data has a stable trend line -> trend line을 fitting한 이후, data로부터 trend를 subtracting(제거)하면 stationary 확보 가능
- if the data isn't trend-stationary, then it might be difference-stationary -> can be removed by differencing
- The simplest way of differencing is to subtract the previous value from each value
- Y(t)-Y(t-1)
ARIMA(Autoregressive Integrated Moving Average)
- Autoregressive Integrated Moving Average: Autoregressive(자기회귀) + Moving Average(이동평균) + Differencing
- Autoregressive Model(AR): 자신의 이전 값 + 확률적 항(불완전하게 예측 가능한 항)에 선형적으로 의존하는 모델 → 자기 자신을 입력 데이터로 하여 스스로를 예측하는 모델(이전 값들을 바탕으로 예측)
- 현재 시점까지 생성한 output을 사용하여 다음 시점의 output 예측 수행
- stationary data에 유리
- Moving Average(MV): 노이즈 제거 → 패턴 파악을 더 쉽게 함
- Autoregressive Model(AR): 자신의 이전 값 + 확률적 항(불완전하게 예측 가능한 항)에 선형적으로 의존하는 모델 → 자기 자신을 입력 데이터로 하여 스스로를 예측하는 모델(이전 값들을 바탕으로 예측)
- 기존의 ARMA 모델: stationary Data에만 적용 가능
- Stationary: 평균, 분산, 자기상관계수 등의 통계적 요소들이 시간의 흐름에 따라 변하지 않음
- common time-series dataset은 non-stationary → 이를 해결하는 것이 differencing(차분)
- non-stationary data는 꾸준한 데이터 변화의 방향성인 trend가 존재하는데, 이를 제거하는 과정이 differencing
- 기존의 ARMA model + differencing = ARIMA → non-stationary data에도 적용 가능
- Two Cases
- Non-seasonal ARIMA model: doesn't require accounting for seasonality in time series data
- predict the future simply based on based on patterns in the past data
- Seasonal ARIMA model: account for seasonality which is regular cycles that affect the time series
- these cycles can be daily, weekly, or monthly
LSTM(Long Short Term Memory)
https://sypdevlog.tistory.com/162
[논문 리뷰] Attention Is All You Need
https://arxiv.org/abs/1706.03762 Attention Is All You Need The dominant sequence transduction models are based on complex recurrent or convolutional neural networks in an encoder-decoder configuration. The best performing models also connect the encoder an
sypdevlog.tistory.com
-> 이전에 정리한 글 참고
- LSTM is a type of RNN(recurrent neural network)
- aimed at mitigating the vanishing gradient problem(기울기 소멸 문제)
- 기존 RNN의 문제인 기울기 소멸 문제 방지 -> 과거의 정보가 점차 소실되는 문제
- 이전 정보를 오랫동안 기억할 수 있는 메모리 셀 소유 -> long sequence data 처리 가능
- input gate, forget gate, output gate를 가져, 셀의 값을 얼마나 기억할지 결정 가능
- sigmoid 함수의 값(0 ~ 1)에 따라서 cell state 정보 기억 여부 결정
- 0: cell state 소실
- 1: cell state 그대로 전달
참고자료
클라우드 데이터센터의 안정적인 운영을 위한 워크로드 예측 모델 비교 연구 | DBpia
정승우, 전민철, 허의남 | 한국정보과학회 학술발표논문집 | 2024.12
www.dbpia.co.kr
Time series - Wikipedia
From Wikipedia, the free encyclopedia Sequence of data points over time Time series: random data plus trend, with best-fit line and different applied filters In mathematics, a time series is a series of data points indexed (or listed or graphed) in time or
en.wikipedia.org
What are ARIMA Models? | IBM
ARIMA stands for Autoregressive Integrated Moving Average and is a technique for time series analysis and forecasting possible future values of a time series.
www.ibm.com
Long short-term memory - Wikipedia
From Wikipedia, the free encyclopedia Type of recurrent neural network architecture The Long Short-Term Memory (LSTM) cell can process data sequentially and keep its hidden state through time. Long short-term memory (LSTM)[1] is a type of recurrent neural
en.wikipedia.org
LSTM(Long short time memory) : 기초 이해
LSTM(Long Short-Term Memory)은 시계열 데이터의 예측, 자연어 처리, 음성 인식, 이미지 분류 등에서 중요한 역할을 하는 모델 중 하나입니다. 이 글에서는 LSTM의 개념, 동작 원리 등에 대해 상세히 알아
ctkim.tistory.com
ARIMA, LSTM을 이용한 시계열 분석 입문
시계열 분석 실습
velog.io
'4학년 1학기 전공 > 캡스톤디자인' 카테고리의 다른 글
[캡스톤 디자인] LSTM(with Example을 통한 동작 과정 이해) (0) | 2025.03.29 |
---|