Train-test split for time series in ML Python - Model Pipeline Trace

Model Pipeline - Train-test split for time series

This pipeline shows how time series data is split into training and testing sets while keeping the order of data. It helps the model learn from past data and test on future data without mixing time order.

Data Flow - 2 Stages

1Original time series data

1000 rows x 1 column→Raw sequential data collected over time→1000 rows x 1 column

[10, 12, 15, 14, 16, 18, 20, ...]

↓

2Train-test split

1000 rows x 1 column→Split data by time order: first 800 rows for training, last 200 rows for testing→Train: 800 rows x 1 column, Test: 200 rows x 1 column

Train: [10, 12, ..., 25], Test: [26, 27, ..., 29]

Training Trace - Epoch by Epoch

Loss
0.5 |****
0.4 |****
0.3 |***
0.2 |**
0.1 |*
    +------------
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.45	0.60	Model starts learning patterns from training data
2	0.35	0.70	Loss decreases, accuracy improves as model fits data better
3	0.28	0.78	Model continues to improve with more training
4	0.22	0.83	Training loss decreases steadily, accuracy rises
5	0.18	0.87	Model converges with good performance on training data

Prediction Trace - 4 Layers

Layer 1: Input test sample

Layer 2: Feature scaling

Layer 3: Model prediction

Layer 4: Inverse scaling

Model Quiz - 3 Questions

Why do we split time series data by order instead of randomly?

A To keep the time order and avoid future data leaking into training

B To make training and testing sets exactly equal in size

C To shuffle data for better randomness

D To reduce the number of features

Key Insight

Splitting time series data by time order ensures the model learns from past data and is tested on future data, preventing data leakage. Training shows steady improvement in loss and accuracy, and scaling inputs helps the model make better predictions.