ARIMA model basics in ML Python - Model Pipeline Trace

Model Pipeline - ARIMA model basics

The ARIMA model helps us predict future points in a series of numbers, like daily temperatures or sales, by learning from past data patterns.

Data Flow - 4 Stages

1Raw time series data

1000 time points x 1 feature→Collect sequential data points over time→1000 time points x 1 feature

Daily sales numbers: [100, 105, 102, 110, ...]

↓

2Differencing

1000 time points x 1 feature→Subtract previous value to remove trend→999 time points x 1 feature

Differences: [5, -3, 8, ...] from sales data

↓

3Model fitting

999 time points x 1 feature→Fit ARIMA(p,d,q) model to data→Trained ARIMA model

ARIMA(1,1,1) fitted to differenced sales

↓

4Forecasting

Trained ARIMA model→Predict future values→Future time points x 1 feature

Predicted sales for next 5 days: [112, 115, 117, 120, 122]

Training Trace - Epoch by Epoch

Loss
0.9 | *
0.8 | *
0.7 |  *
0.6 |   *
0.5 |    *
0.4 |     **
    +------------
     1 2 3 4 5 Epochs

Epoch	Loss ↓	Accuracy ↑	Observation
1	0.85	N/A	Initial model fit with high error
2	0.65	N/A	Model parameters adjusted, error decreased
3	0.50	N/A	Better fit, loss steadily decreasing
4	0.45	N/A	Model converging, loss improvement slowing
5	0.43	N/A	Final model fit with stable low error

Prediction Trace - 4 Layers

Layer 1: Input recent time points

Layer 2: Apply differencing

Layer 3: Use AR and MA components

Layer 4: Invert differencing

Model Quiz - 3 Questions

What is the purpose of differencing in ARIMA?

A To increase the number of data points

B To add noise to the data

C To remove trend and make data stable

D To split data into training and testing sets

Key Insight

ARIMA models work by making data stable through differencing, then learning patterns in changes to predict future values. Watching loss decrease during training shows the model is improving its predictions.