Imputation Methods Outperform Missing-Indicator for Data Missing Completely at Random

Barata, António Pereira; Takes, Frank W.; Herik, Jaap van den; Veenman, Cor J.

doi:10.1109/icdmw.2019.00066

Cited by 9 publications

(4 citation statements)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Barata et al (7) discussed the use of imputation methods to handle missing data that are missing completely at random (MCAR). The authors compared the performance of imputation methods to the commonly used approach of using missing indicators.…”

Section: Related Workmentioning

confidence: 99%

A Novel Composite DTERM-MDI Model for the Recovery of Missing Data in Internet of Medical Things

Punitha,

Sathiaseelan

2023

IJST

View full text Add to dashboard Cite

Background/Objectives: The Internet of Things (IoT) relies on consistent data delivery, crucial for maintaining service quality. However, challenges like connection issues, external threats, and sensor malfunctions can lead to data insufficiency, affecting IoT applications. Addressing the problem of missing data in the vast streams of IoT-generated data is essential. This paper introduces a novel Composite DTERM Missing Data Imputation (MDI) model for the Internet of Medical Things (IoMT) aimed at robustly recovering missing data. Methods: The Composite DTERM-MDI model comprises three phases: Dual Strategy based Missing Data Imputation (DS-MDI), Two Tier Missing Data Imputation (TT-MDI), and Ensemble Regression Model (ERM) for progressive imputation of Missing Not at Random (MNAR) type data. The effectiveness of the Composite DTERM-MDI model is demonstrated using the cStick dataset, achieving significantly improved accuracy, precision, recall, and F-measure compared to the original dataset. Additionally, a comparison of the Composite DTERM-MDI model with two standard imputation techniques, MICE and SICE, is performed using the UCI car dataset. Findings: Experimental results showcase the superiority of the Composite DTERM-MDI model-based imputed cStick dataset, with accuracy at 99.12%, precision at 99.98%, recall at 98.53%, and F-measure at 98.89%, outperforming the original cStick dataset. The study highlights the Composite DTERM-MDI model's efficiency and accuracy in addressing missing data challenges in IoMT, which is vital for informed medical decision-making. Novelty: Furthermore, a comparison of the Composite DTERM-MDI model with MICE and SICE using the UCI car dataset evaluates accuracy and F-measure across four classification algorithms (PMM, POLYREG, CART, and LDA). The Composite DTERM-MDI model achieves accuracy rates of 91.08%, 81.58%, 97.74%, and 97.74%, along with F-measures of 84.97%, 89.58%, 98.71%, and 99.17% for PMM, POLYREG, CART, and LDA, respectively. This comparison demonstrates the model's performance against established imputation techniques in a different context. https://www.indjst.org/

show abstract

Section: Related Workmentioning

confidence: 99%

A Novel Composite DTERM-MDI Model for the Recovery of Missing Data in Internet of Medical Things

Punitha,

Sathiaseelan

2023

IJST

View full text Add to dashboard Cite

show abstract

“…Vamplew and Adams (1992) and Ng and Yusoff (2011) only use a single dataset with randomly removed values, and base their evaluation on the performance of a single algorithm (respectively a neural network and linear regression). Pereira Barata et al (2019) use three classification algorithms and 22 datasets, but again with randomly removed values, explicitly assuming a MCAR context. They conclude that imputation outperforms missing-indicators, but the comparison is not like-for-like, since it involves several forms of imputation but only combines indicator attributes with zero imputation.…”

Section: Previous Experimentsmentioning

confidence: 99%

No imputation without representation

Lenz¹,

Peralta²,

Cornelis³

2022

Preprint

View full text Add to dashboard Cite

By filling in missing values in datasets, imputation allows these datasets to be used with algorithms that cannot handle missing values by themselves. However, missing values may in principle contribute useful information that is lost through imputation. The missingindicator approach can be used in combination with imputation to instead represent this information as a part of the dataset. There are several theoretical considerations why missing-indicators may or may not be beneficial, but there has not been any large-scale practical experiment on real-life datasets to test this question for machine learning predictions. We perform this experiment for three imputation strategies and a range of different classification algorithms, on the basis of twenty real-life datasets. We find that on these datasets, missing-indicators generally increase classification performance. In addition, we find no evidence for most algorithms that nearest neighbour and iterative imputation lead to better performance than simple mean/mode imputation. Therefore, we recommend the use of missing-indicators with mean/mode imputation as a safe default, with the caveat that for decision trees, pruning is necessary to prevent overfitting. In a follow-up experiment, we determine attribute-specific missingness thresholds for each classifier above which missing-indicators are more likely than not to increase classification performance, and observe that these thresholds are much lower for categorical than for numerical attributes. Finally, we argue that mean imputation of numerical attributes may preserve some of the information from missing values, and we show that in the absence of missing-indicators, it can similarly be useful to apply mean imputation to one-hot encoded categorical attributes instead of mode imputation.

show abstract

“…In this manner, a value of 0 is placed where values are missing ('NaN') and an additional column is added representing the missingness of each attribute (Figure 2). Given classifier robustness, the results obtained in performance should not alter significantly by using other imputation methods [18].…”

Section: Datamentioning

confidence: 99%

Large-scale machine learning for business sector prediction

Angenent

Barata

Takes

2020

Proceedings of the 35th Annual ACM Symposium on Applied Computing

View full text Add to dashboard Cite

In this study we use machine learning to perform explainable business sector prediction from financial statements. Financial statements are a valuable source of information on the financial state and performance of firms. Recently, large-scale data on financial statements has become available in the form of open data sets. Previous work on such data mainly focused on predicting fraud and bankruptcy. In this paper we devise a model for business sector prediction, which has several valuable applications, including automated error and fraud detection. In addition, such a predictive model may help in completing similar datasets with missing sector information. The proposed method employs a supervised learning approach based on random forests that addresses business sector prediction as a classification task. Using a dataset from the Netherlands Chamber of Commerce, containing over 1.5 million financial statements from Dutch companies, we created an adequately-performing model for business sector prediction. By assessing which features are instrumental in the final classification model, we found that a small number of attributes is crucial for predicting the majority of business sectors. Interestingly, in some cases the presence or absence of a feature was more important than the value itself. The resulting insights may also prove useful in accounting, where the relation between financial statements and characteristics of the company is a frequently studied topic.

show abstract

Imputation Methods Outperform Missing-Indicator for Data Missing Completely at Random

Cited by 9 publications

References 29 publications

A Novel Composite DTERM-MDI Model for the Recovery of Missing Data in Internet of Medical Things

A Novel Composite DTERM-MDI Model for the Recovery of Missing Data in Internet of Medical Things

No imputation without representation

Large-scale machine learning for business sector prediction

Contact Info

Product

Resources

About