Skip to main content
Skip table of contents

Root Cause Analysis (RCA)

Introduction

Root Cause Analysis (RCA) is a vital tool to obtain insights into the driving factors influencing a forrecast. By initiating RCA on a completed forecasting job, users gain a deeper understanding of the elements contributing to the forecast and how different models within the Model Zoo shape the final output. RCA is particularly useful for understanding structural changes in normal behaviour models, helping users identify the specific terms and predictors responsible for anomalies.

Output Overview

When RCA is conducted, it generates a detailed output that includes various terms, forecast values (yhat), and predictors for each timestamp. The output is typically presented in a tabular format with the following columns:

  • Timestamp: The specific time point of the forecast.

  • Terms (term_1, term_2, ..., term_N): The individual components or terms of the model that contribute to the forecast.

  • Forecast Values (yhat_1, yhat_2, ..., yhat_N): The forecasted values at each step as the model builds up.

  • Predictors (predictor_1, predictor_2, ..., predictor_N): The variables or features that influence the forecast.

timestamp

term_1

term_2

term_3

term_N

yhat_1

yhat_2

yhat_3

yhat_N

predictor_1

predictor_2

predictor_3

predictor_N

2014-10-25T04:00:00.0

2546

900

943.05

624

1943

1984

1987

3296

1443

1984

1287

1396

2014-10-26T04:00:00.0

2451

5000

5409.6

234

2195

2104

2089

3123

2943

1584

2987

1496

2014-10-27T04:00:00.0

2103

200

65363.4

123

2211

2190

2168

2762

2142

1984

2987

996

2014-10-28T04:00:00.0

2301

100

543.5

545

2189

2154

2167

4153

643

1984

1987

1996

2014-10-29T04:00:00.0

2225

432

983

321

2567

2592

2598

3112

1143

1484

1987

1996

2014-10-30T04:00:00.0

2155

4355

1235.6

134

2532

2490

2487

4123

4943

1984

1987

1996

This output allows users to dissect the forecast into its constituent parts and understand how each component contributes to the overall prediction.

Interpreting RCA Results

RCA results are interpreted through three distinct views: Nominal Term View, Relative Term View, and Predictor View. Each view offers a unique perspective on the forecast and the underlying model construction.

Nominal Term View

  • Description: Displays the exact contribution of each model term to the final forecast value. The sum of all terms equals the forecast value at any given timestamp.

  • Purpose: Helps users understand how each term contributes to the forecast, offering clarity on the specific impact of individual model components.

  • Application: Use this view to dissect the forecast into its components and compare the contributions of terms across different models. This is particularly useful when diagnosing discrepancies in forecast accuracy.

Relative Term View

  • Description: Shows how the forecast builds as each term is sequentially added to the model, illustrating the cumulative effect of terms.

  • Purpose: Highlights the incremental importance of each term in shaping the final forecast, helping users visualize how the model's accuracy improves as more terms are included.

  • Application: Use this view to identify which specific term might be introducing errors by analyzing how the forecast value changes with each added term.

Predictor View

  • Description: Aggregates the effects of each predictor across all terms in the model, showing their total contribution to the forecast.

  • Purpose: Simplifies the analysis by focusing on the overall influence of each predictor, rather than breaking down individual terms.

  • Application: Use this view to evaluate the significance of different input variables (predictors) and their overall impact on the forecast. This is particularly useful for understanding which predictors drive the forecast and should be adjusted or emphasized.

JavaScript errors detected

Please note, these errors can depend on your browser setup.

If this problem persists, please contact our support.