Paper Number

1273

Paper Type

Short

Description

Algorithmic forecasts outperform human forecasts by 10% on average. State-of-the-art machine learning (ML) algorithms have further expanded this discrepancy. Because a variety of other activities rely on them, sales forecasting is critical to a company's profitability. However, individuals are hesitant to use ML forecasts. To overcome this algorithm aversion, explainable artificial intelligence (XAI) can be a solution by making ML systems more comprehensible by providing explanations. However, current XAI techniques are incomprehensible for laymen, as they impose too much cognitive load. We contribute to this research gap by investigating the effectiveness in terms of forecast accuracy of two example-based explanation approaches. We conduct an online experiment based on a two-by-two between-subjects design with factual and counterfactual examples as experimental factors. A control group has access to ML predictions, but not to explanations. We report results of this study: While factual explanations significantly improved participants’ decision quality, counterfactual explanations did not.

Comments

09-HCI

Share

COinS
 
Dec 12th, 12:00 AM

Effectiveness of Example-Based Explanations to Improve Human Decision Quality in Machine Learning Forecasting Systems

Algorithmic forecasts outperform human forecasts by 10% on average. State-of-the-art machine learning (ML) algorithms have further expanded this discrepancy. Because a variety of other activities rely on them, sales forecasting is critical to a company's profitability. However, individuals are hesitant to use ML forecasts. To overcome this algorithm aversion, explainable artificial intelligence (XAI) can be a solution by making ML systems more comprehensible by providing explanations. However, current XAI techniques are incomprehensible for laymen, as they impose too much cognitive load. We contribute to this research gap by investigating the effectiveness in terms of forecast accuracy of two example-based explanation approaches. We conduct an online experiment based on a two-by-two between-subjects design with factual and counterfactual examples as experimental factors. A control group has access to ML predictions, but not to explanations. We report results of this study: While factual explanations significantly improved participants’ decision quality, counterfactual explanations did not.

When commenting on articles, please be friendly, welcoming, respectful and abide by the AIS eLibrary Discussion Thread Code of Conduct posted here.