The selection of useful, informative, and meaningful features is a key prerequisite for the successful application of machine learning in practice, especially in knowledge-intense domains like decision support. Here, the task of feature selection, or ranking features by importance, can, in principle, be solved automatically in a data-driven way but also supported by expert knowledge. Besides, one may of course, conceive a combined approach, in which a learning algorithm closely interacts with a human expert. In any case, finding an optimal approach requires a basic understanding of human capabilities in judging the importance of features compared to those of a learning algorithm. Hereto, we conducted a case study in the medical domain, comparing feature rankings based on human judgment to rankings automatically derived from data. The quality of a ranking is determined by the performance of a decision list processing features in the order specified by the ranking, more specifically by so-called probabilistic scoring systems.
CEUR ceur-ws.org
CEUR Workshop Proceedings
That said, decision models constructed in a data-driven way will not be accepted by human experts 3[] — and hence not be used in practice — unless these models are comprehensible, meaningful, and interpretable. In this regard, the selection and prioritization of decision criteria, orfeatures in machine learning jargon, appears to be of major importance: The features on which a decision is based need to be semantically meaningful; features deemed relevant by the expert should be included in the model, while irrelevant features should be omitted.
Needless to say, these properties are not necessarily guaranteed when selecting features in a purely data-driven way. As another extreme, one may think of letting the human expert preselect the features by hand. For various reasons, however, this might be suboptimal either, for example, because the expert might be subjectively biased, or her knowledge might not be perfect. Presumably, the best approach is somewhere in-between, namheylbyr,id in the sense that the human expert and the machine learning algorithm select features jointly in the course of an interactive process. Either way, these considerations beg an essential question: How capable are human experts in selecting the most important features or in ranking features in descending order of importance, and how do human experts compare to ML algorithms selecting features in a data-driven mann4,e5r, [6]?
This is the question addressed by the current paper. We conducted a case study in the medical domain, comparing feature (importance) rankings based on human judgment to feature rankings derived from data. The quality of a ranking is determined by the performance of a decision list processing features in the order specified by the ranking. In a decision list, features are considered incrementally, one by one. In each stage of the process, there are two options: either a final decision is made based on the feature values seen so far, or the process is continued by observing the next feature. Features should be ranked in decreasing order of importance to make well-informed decisions as quickly as possible. We implement this approach with so-called scoring systems, specifically appealing from an interpretability perspective and commonly used in the medical domain7[, 8].
Previous research suggests that data-driven methods generally surpass knowledge-driven methods in performance, though these findings are not entirely unambiguous. Our study contributes to resolving this continuing debate and extends the current literature by assessing these methods within the context of interpretable machine learning models. In high-stakes environments such as in the medical domain, tchoenstructor of the decision model can be a significant factor for decision-makers, influencing their trust and reliance on the system. Consequently, evaluating the quality of various feature selection methods on such models is vital.
Our study shows that while data-driven feature ranking exhibits superior performance in identifying patterns unseen by human actors, the risk of overfitting, especially in small or biased datasets, necessitates the incorporation of human judgment for optimal results. We suggest an interactive, co-constructive approach, merging human expertise with algorithmic analytics, as a potential solution to ofset overfitting efects while enhancing user acceptance of decision models. We encourage future research to leverage our findings, specifically targeting the inclusion of more domain professionals in the dataset, to further enrich and generalize these insights across various fields. 2. Data- and Knowledge-Driven Feature Selection In the realm of supervised machine learning, most algorithms assume a representation of data objects (instances) in terms of feature vectors, which means that each object is specified by its values on a predefined number of features, also known as independent variables, dimensions, or inputs. The latter are supposed to carry important information for predicting the outcome or target variabl9e].[ Careful feature selection is a crucial step in the modeling process and a key prerequisite for learning accurate predic1t0o]r.sS[electing a manageable number of meaningful features also facilitates interpretability and explaina6b].ility [
Feature selection has been researched intensively in the past, with a specific focus on
datadriven approaches. Here, an algorithm autonomously ranks or selects features based on the
properties of the data. In contrast, knowledge-driven approaches determine a feature subset
through literature revie1w1,[12, 13] or by consulting domain expert4s, [14]. Interactive
machine learning fosters a combination of these approac1h5e]s. F[or instance, experts might
underscore highly relevant observations and features that a data-driven algorithm can
subsequently focus on1[6]. Alternatively, experts might vote on diferent feature subsets, indirectly
revealing their subjective preferenc1e7s].[ It is also possible to aggregate multiple selection
and ranking methods into a single approa1c8h, [
Choosing the optimal method for a specific dataset and problem domain is inherently
challenging. Guyon and Elisseef [21] and Li et al. 6[] advocate for including domain knowledge
in the selection process. Conversely, Filippova et a5l]. fin[d human intervention to be less
beneficial than expected, while McKay2[
A so-calledscoring system is a simple decision model that checks a set of features, adds (or
subtracts) a certain number of points to a total score for each feature that is satisfied, and finally
makes a decision by comparing the total score to a threshold. Scoring systems have a long
history of active use in safety-critical domains such as health2c3a]raen[d justice2[
Hanselle et al.2[
The learning algorithm introduced in Hanselle et2a5]l.c[onstructs PSLs incrementally in a greedy manner. Starting with the empty list, one additional feature with a corresponding score (taken from a predefined set of scores) is added to the list in each stage. To this end, each feature/score pair is tentatively added as a candidate, and the resulting model is evaluated in terms of theexpected entropy as performance measure: = ∑ ⋅ ((̂ ) ) , ∈Σ (1) where Σ is the set of total scores that can be produced by the decisi on= l|i s|t, is the total number of training examples, a n dthe number of training examples with total s co.re Moreover,(̂ ) is the estimated probability of the positive class given total s,caonrde is the Shannon entropy () = − ⋅
log() − (1 − ) log(1 − ) .
The feature/score combination leading to the highest performance is eventually added to the list, and the algorithm proceeds to the next stage (unless all features are used or the gain in terms of expected entropy is negative). The probabili(t̂)ies are estimated in terms of relative frequencies, rectified by isotonic regression to guarantee monotonicity (the probability of the positive class increases with an increasing total score).
Note that the expected entro1p)yis(a meaningful measure of informedness at every stage of the decision process: The information provided by the prediction of a probability dist rîbution is quantified in terms of Shannon entropy, which is an established measure of information, and weighted by the (estimated) probability that this prediction is delivered.
The PSL produced by the above algorithm also suggests a ranking of features in the sense that features appearing earlier in the list seem to be more important in terms of performance than features queried only later on (or possibly not at all, if a decision is made before). With a straightforward modification, the algorithm can also be used to learn scoring systems for a predefined ranking of features: In each stage, it then adopts the corresponding feature and only optimizes over the set of possible scores, instead of optimizing over all features/score pairs.
In the following, we compare PSLs constructed solely in a data-driven fashion to PSLs in which the evaluated features are ordered according to human choices.
We employed a non-public medical dataset, based on the work of Hüfner e2t6a].l.A[ minor deviation from the original dataset in our study pertains to the exclusion of a single observation that contained a missing value. Consequently, our dataset has a total of 696 patient observations.
According to the medical tests conducted in the original study, 633 patients (90.95%), tested negative for COVID-19. This dataset is comprised of 11 binary features, which, apart from information regarding patient contact with an infected individual, include all patient symptoms. Figure1 shows all features, their respective distributions across the entire dataset, and the distributions for both positive and negative cases. While our dataset does not include additional demographic information, Hüfner et a2l6.][ state in their study that 51.1% of the patients were female and the average age was 55.2 years.
Figure2 shows the correlation between all features. Quite remarkably, the feature “Contact with an infected person” is negatively correlated to the target variable. Intuitively, contact with an infected person and the associated risk of exposure to the virus should have a positive correlation with an infection. One possible explanation for this peculiarity might be, that people who know that they had contact with an infected person may have higher awareness 0:6 e c n le0:4 a rev P 0:2 0:0 Class Overall Positive Class Negative Class Contact with infected person
Cough Digestive proFbaletmigsue Fever
and hence be tempted to ask for a medical examination more quickly, even when showing no clear symptoms. This trend is further observable in the first column of the heatmap, where the correlations with symptoms such as respiratory issues and fever also exhibit a negative association.
In our experimental evaluation, we use PSLs constructed in five diferent manners. First, we consider PSLs derived from training data using the algorithm described in S3e.cTthioense are calledPSL.
Second, we compare them against PSLs built from expert input, specifically the origCionvaidl Score system proposed by Hüfner et al2.6[] (Expert-PSL). The Covid Score was compiled as a consensus of medical experts. It was evaluated on the proposed dataset, however, it has not been used in the process of deriving the score. Note Etxhpaetrt-PSL is a probabilistic scoring list and thus conceptually diferent from the original scoring system, which always evaluates the entire feature set and uses a constant thresh5oalsdaofdecision rule.
Two further approaches are derived based on a recent incentivized behavioral experiment conducted by Kornowicz and Thomme2s7[]. In this study, 234 subjects, recruited from the Prolific.co 1 platform, were requested to rank features based on their perceived importance for the classification task. Despite these subjects lacking specific medical field expertise, it remains plausible that the aggregate of their rankings might approximate the quality of expert opinions, as suggested by research in the field of expert elicitati2o8n,2[9, 30]. We primarily utilized the rankings generated individually by subjecStusb(ject-PSL), along with a method of consensus ranking referred to as Behavioral AggregatSiuobnje(ctBA-PSL). For this method, 90 subjects were grouped into sets of three to agree upon a collective ranking. As there are no specified scores attached to the latter, we chose the scores associated with the features in the same greedy, data-driven manner as the first approach to allow for a fair comparison.
Lastly, as a baseline, we consider PSLs constructed from random feature permutations, for which the scores have been chosen in the same mannRearn(dom-PSL). We chose = {±1, ±2, ±3} as the set of possible scores for all methods except the expert method. The expert method’s scores are taken from the scoring system by Hüfner e2t6a]la.n[d hence constrained to = {+1, +2, +3} . An overview of the considered constructions is depicted in 2T.able
We evaluated the individual PSLs in terms of a Monte Carlo cross-validation (MCCV) with 10 repetitions. In each repetition, we use a fraction of two-thirds of the available data as training data and one-third as test data. We report the expected entropy as a neutral measure of informativeness at each stage of the decision model in order to compare the approaches. Additionally, we evaluate the decision models in terms of expected loss minimization. In the domain of medical decision-making, it is common that a false negative prediction, i.e., not isolating and treating a COVID-19-infected patient, has far more severe consequences than a false positive. To capture this, we employ an asymmetric loss function that assigns a 1loss of to false positives and a loss o≫f 1 to false negatives. Given the PSLs probabilistic prediction ̂ for the positive class, the risk-minimizing decision is =̂ { 1 if 1 − <̂ ⋅ ̂ 0 otherwise , and the (estimated) expected loss itsel f(b)̂y= min{1 − , ̂ ⋅ } ̂ . For the experiments, we chose ≔ 10 , i.e., penalizing false negatives ten times as much as false positives.
In the following, we compare the five diferent PSL constructions against each other. Figure 3 shows the mean expected entropy and expected loss of the PSLs for each stage, i.e., after evaluating the stated number of features.
We observe thaPtSL achieves the best mean expected entropy throughout all stages. The Subject-PSL andSubjectBA-PSL constructions perform very similar. Up until st3a, gtehey exhibit a higher mean expected entropy thanRtahnedom-PSL baseline before consistently outperforming it as of stage5. The Expert-PSL construction also performs worse than the random baseline within the first stages, even deteriorating when evaluating the first two features, both in terms of expected entropy as well as expected loss. This is due to the fact that the first two features selected bEyxpert-PSL are “Contact w/ inf. person” and “Respiratory symptom”. 1:0
6 Stage Expert-PSL 8 10 0 2 4 8
10 6
Stage
As already discussed in Sectio4n.1, the “Contact w/ inf. person” is negatively correlated with the target “SARS-CoV-2 positive” and the respiratory symptom is only weakly positively correlated to it. These two features both receive a sco+re3 oinf theExpert-PSL construction, yielding poor performances early on and even deteriorating over the performance a0tinstwahgiech no feature is considered. The fact that these two features, which seem quite indicative for the human eye, do not have a strong positive influence on the outcome remains undiscovered for the experts. Here, the data-driven approPaScLhtakes advantage of having access to training data, placing it on average at a ran9k. of k11 n a 9 R r 7 e u t e 5 a F an 3 eM1 e 3 r co 2 S re 1 u ta 0 e F 1 n ea 2 M3
Contact w/ inf. person Cough
FeverHeadacheLimb pLaRoinsesspoifrastmoreyllsymptom Sni eSsore throat PSL
Expert-PSL
Subject-PSL
SubjectBA-PSL
Figure4 shows an overview of the average ranks and scores of all features across the considered methods. For many features, the average ranks of the diferent approaches are quite similar, with the exception of the “Fatigue” and the “Contact w/ inf. person”. Since the scores are optimized to the data in all approaches except foErxtpheert-PSL, the scores are really similar. This holds true regardless of the average rank of the feature. Note that the expert scores are selected according to Hüfner et2a6l]., [constraining them to only positive scores. Reducing available training data As discussed in the previous section, the data-driven approachPSL manages to unveil specifics from the data that are not taken into account by human actors. To make this feasible, it makes use of training data whose availability is a necessary condition for applying such methods. To investigate how much the data-driven approaches are dependent on the availability of data, we restricted 2t0h%emofttohe original training data by drawing subsamples from the original data without replacement and repeated the experiments 10 times. Figur5eshows the expected entropy of the diferent PSLs when training them on these reduced training datasets.
0:45
6 Stage Expert-PSL 8 10 2 4
6 Stage 8
10 Subject-PSL
SubjectBA-PSL
Random-PSL
We observe that thPeSL is outperformed from stage 7 on by tEhxepert-PSL and also by the Subject-PSL andSubjectBA-PSL as of stage 9 in terms of expected entropy. When it comes to the expected losPs,SL is already beaten bEyxpert-PSL at stage 3 and thSeubject-PSL andSubjectBA-PSL methods at stage 7. In the end, even tRhaendom-PSL baseline exhibits a slightly lower mean expected error thanPtShLe. As expected, data-driven approaches become less reliable once access to data is restricted. In such scenarios, human expertise and common sense achieve better results than automated methods.
This paper has explored the comparative efectiveness of humans and algorithms in feature ranking for decision support. A case study in the medical domain was conducted, in which we compared feature rankings based on human judgment to rankings automatically derived from data. It was observed that the data-driven approach can identify patterns and specifics that remained hidden from human actors, leading to better performances in our experimental evaluation. On the other hand, feature rankings solely derived in an algorithmic manner bear the risk of being overfitted to the available training data, resulting in poor generalization performance. This becomes especially important when training datasets are small or significantly biased. In this case, human knowledge and common sense may be a good countermeasure to compensate for such efects.
An interactive feature ranking procedure that combines the strengths of human and
datadriven approaches constitutes an interesting direction for future work. Harnessing the benefits
of human expertise and computational analytics in a co-constructive approach potentially
leads to more accurate decision models while mitigating the risk of overfitting. Additionally,
including humans in the learning procedure may also increase the practitioner’s acceptance of
the obtained decision model, as purely algorithmically constructed models are often faced with
distrust3[
As machine learning-based decision support systems continue to gain traction, our findings ofer valuable insights to researchers in this emerging field. Future research eforts could potentially build upon and generalize our findings by employing diferent datasets and extending the scope to various domains. One of the key strengths of our dataset is the high volume of human rankings; however, these subjects notably lack significant domain experience, with the exception of the utilization of Ctohveid Score system of Hüfner et al. 2[6]. While the recruitment of a larger number of domain professionals presents a challenge, pursuing this could undoubtedly yield more insightful findings in future research.
We gratefully acknowledge funding by the German Research Foundation (Deutsche Forschungsgemeinschaft, DFG): TRR 318/1 2021 – 438445824. [6] J. Li, K. Cheng, S. Wang, F. Morstatter, R. P. Trevino, J. Tang, H. Liu, Feature selection: A data perspective, ACM Computing Surveys 50 (2017) 94:1–94:45. d1o0i.: 1145/3136625. [7] A. G. Rapsang, D. C. Shyam, Scoring systems in the intensive care unit: A compendium, Indian Journal of Critical Care Medicine : Peer-reviewed, Oficial Publication of Indian Society of Critical Care Medicine 18 (2014) 220–228. d10o.i:4103/0972-5229.130573. [8] B. Ustun, C. Rudin, Supersparse linear integer models for optimized medical scoring systems, Machine Learning 102 (2016) 349–391. do1i0: .1007/s10994-015-5528-6. [9] G. James, D. Witten, T. Hastie, R. Tibshirani, An introduction to statistical learning, volume 112, Springer, 2013. [10] S. Studer, T. B. Bui, C. Drescher, A. Hanuschkin, L. Winkler, S. Peters, K.-R. Müller, Towards crisp-ml(q): A machine learning process model with quality assurance methodology, Machine Learning and Knowledge Extraction 3 (2021) 392–4131.d0.o3i:390/make3020020. [11] D. C. Corrales, E. Lasso, A. Ledezma, J. C. Corrales, Feature selection for classification tasks: Expert knowledge or traditional methods?, Journal of Intelligent & Fuzzy Systems 34 (2018) 2825–2835. doi:10.3233/JIFS-169470. [12] J. Nahar, T. Imam, K. S. Tickle, Y.-P. P. Chen, Computational intelligence for heart disease diagnosis: A medical knowledge driven approach, Expert Systems with Applications 40 (2013) 96–104. doi:10.1016/j.eswa.2012.07.032. [13] J. Wang, J. Oh, H. Wang, J. Wiens, Learning credible models, in: Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD ’18, Association for Computing Machinery, New York, NY, USA, 2018, p. 2417–2426. URL: https://doi.org/10.1145/3219819.322007.0doi:10.1145/3219819.3220070. [14] S. Moro, P. Cortez, P. Rita, A divide-and-conquer strategy using feature relevance and expert knowledge for enhancing a data mining approach to bank telemarketing, Expert Systems 35 (2018) e12253. doi1:0.1111/exsy.12253. [15] A. Holzinger, Interactive machine learning for health informatics: when do we need the human-in-the-loop?, Brain Informatics 3 (2016) 119–131. 1d0o.i1:007/ s40708-016-0042-6. [16] A. H. C. Correia, F. Lecue, Human-in-the-loop feature selection, Proceedings of the AAAI Conference on Artificial Intelligence 33 (2019) 2438–2445. d1o0i:.1609/aaai.v33i01. 33012438. [17] F. Bianchi, L. Piroddi, A. Bemporad, G. Halasz, M. Villani, D. Piga, Active preference-based optimization for human-in-the-loop feature selection, European Journal of Control 66 (2022) 100647. doi:10.1016/j.ejcon.2022.100647. [18] V. Bolón-Canedo, A. Alonso-Betanzos, Ensembles for feature selection: A review and future trends, Information Fusion 52 (2019) 1–12.1d0o.i1:016/j.inffus.2018.11.008. [19] D. J. Dittman, T. M. Khoshgoftaar, R. Wald, A. Napolitano, Classification performance of rank aggregation techniques for ensemble gene selection, in: The twenty-sixth international FLAIRS conference, 2013. [20] R. Wald, T. M. Khoshgoftaar, D. Dittman, W. Awada, A. Napolitano, An extensive comparison of feature ranking aggregation techniques in bioinformatics, in: 2012 IEEE 13th International Conference on Information Reuse & Integration (IRI), 2012, p. 377–384. doi:10.1109/IRI.2012.6303034. [21] I. Guyon, A. Elisseef, An introduction to variable and feature selection, Journal of machine learning research 3 (2003) 1157–1182. [22] S. McKay, When 4≈ 10,000: The power of social science knowledge in predictive performance, Socius 5 (2019) 2378023118811774. [23] A. Six, B. Backus, J. Kelder, Chest pain in the emergency room: value of the heart score,
Netherlands Heart Journal 16 (2008) 191–196. [24] C. Wang, B. Han, B. Patel, C. Rudin, In pursuit of interpretable, fair and accurate machine learning for criminal recidivism prediction, Journal of Quantitative Criminology (2022) 1–63. [25] J. Hanselle, F. Fürnkranz, E. Hüllermeier, Probabilistic scoring lists for interpretable machine learning, in: Proc. DS, 23rd Int. Conference on Discovery Science, Springer, Porto, Portugal, 2023. [26] A. Hüfner, D. Kiefl, M. Baacke, R. Zöllner, E. Loza Mencía, O. Schellein, N. Avan, S. Pemmerl, Risikostratifizierung durch implementierung und evaluation eines covid-19scores, Medizinische Klinik - Intensivmedizin und Notfallmedizin 115 (2020) 132–138. doi:10.1007/s00063-020-00754-4. [27] J. Kornowicz, K. Thommes, Aggregating human domain knowledge for feature ranking, in: H. Degen, S. Ntoa (Eds.), Artificial Intelligence in HCI, Lecture Notes in Computer Science, Springer Nature Switzerland, Cham, 2023, p. 98–114. d1o0i.:1007/978-3-031-35891-3_ 7. [28] D. Önkal, J. F. Yates, C. Simga-Mugan, Ş. Öztin, Professional vs. amateur judgment accuracy: The case of foreign exchange rates, Organizational Behavior and Human Decision Processes 91 (2003) 169–185. do1i:0.1016/S0749-5978(03)00058-X. [29] M. Nofer, Are Crowds on the Internet Wiser than Experts? – The Case of a Stock Prediction Community, Springer Fachmedien, Wiesbaden, 2015, p. 27–61. URLh:ttps://doi.org/10. 1007/978-3-658-09508-6_3. doi:10.1007/978-3-658-09508-6_3. [30] E. Vul, H. Pashler, Measuring the crowd within: Probabilistic representations within individuals, Psychological Science 19 (2008) 645–647. do10i:.1111/j.1467-9280.2008. 02136.x. [31] H. Mahmud, A. K. M. N. Islam, S. I. Ahmed, K. Smolander, What influences algorithmic decision-making? a systematic literature review on algorithm aversion, Technological Forecasting and Social Change 175 (2022) 121390. d1o0i:.1016/j.techfore.2021.121390.