Time series components (trend, seasonality) in ML Python - Model Pipeline Trace

Model Pipeline - Time series components (trend, seasonality)

This pipeline breaks down a time series into its main parts: trend and seasonality. It helps us understand how data changes over time and repeats in patterns.

Data Flow - 4 Stages

1Raw time series data

365 days x 1 column→Collect daily values (e.g., temperature or sales)→365 days x 1 column

Day 1: 20, Day 2: 22, Day 3: 21, ..., Day 365: 19

↓

2Detrend data

365 days x 1 column→Remove long-term trend using moving average→365 days x 1 column

Original: 20, 22, 21, ...; Detrended: -1, 0, -0.5, ...

↓

3Extract seasonality

365 days x 1 column (detrended)→Calculate repeating patterns (e.g., weekly or yearly)→365 days x 1 column

Seasonal pattern: +2 on weekends, -1 on weekdays

↓

4Residual calculation

365 days x 1 column→Subtract trend and seasonality from original data→365 days x 1 column

Residuals: small random noise values around 0

Training Trace - Epoch by Epoch

Loss
0.5 |****
0.4 |*** 
0.3 |**  
0.2 |*   
0.1 |    
    +----
     1 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.45	N/A	Initial decomposition with high error in residuals
2	0.30	N/A	Trend smoothing improved, residuals smaller
3	0.20	N/A	Seasonality captured better, residual noise reduced
4	0.15	N/A	Stable decomposition, residuals close to random noise
5	0.12	N/A	Final model captures trend and seasonality well

Prediction Trace - 4 Layers

Layer 1: Input raw value

Layer 2: Remove trend

Layer 3: Remove seasonality

Layer 4: Calculate residual

Model Quiz - 3 Questions

What does the trend component represent in a time series?

A The long-term increase or decrease in data

B Random noise in the data

C Daily repeating patterns

D Missing values in the data

Key Insight

Breaking a time series into trend and seasonality helps us understand and predict data better by separating long-term changes from repeating patterns.