Downloads: 3 | Views: 59 | Weekly Hits: ⮙2 | Monthly Hits: ⮙3
Informative Article | Science and Technology | India | Volume 10 Issue 3, March 2021
Accuracy vs. Interpretability: Balancing Trade - Offs in Forecasting Models
Abstract: As data scientists delve into the realm of forecasting models, a crucial consideration emerges - the delicate trade - off between accuracy and interpretability. This paper explores the intrinsic relationship between these two metrics, recognizing that heightened accuracy often begets decreased interpretability and vice versa. The discussion extends to the broader notion of accuracy, expressed as a percentage and evaluated through complementary metrics like accuracy and error rates. This section emphasizes the importance of accuracy in gauging the overall correctness of forecasting models, particularly applicable in classification scenarios. On the flip side, interpretability surfaces as a crucial facet in forecasting, denoting the ease with which humans can decipher a model's decision - making processes. The paper then navigates the strategies for balancing the accuracy - interpretability trade - off. Linear models and decision trees emerge as interpretable alternatives, while ensemble models and deep learning architecture promise heightened accuracy at the cost of interpretability. Exponential smoothing models present an intriguing middle ground, offering a balance between accuracy and interpretability in time series forecasting.
Keywords: forecasting, interpretability, linear models, ensemble, regression, machine learning, deep learning
Edition: Volume 10 Issue 3, March 2021,
Pages: 1964 - 1966