{"id":"e8dccc5f-a5ac-4690-8117-8529ac1b7c81","arxiv_id":"2501.10980","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":1.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"A disjointed review-style preprint that asserts chi-squared feature selection improves lung cancer detection accuracy but provides no experimental evidence for its central claim.","lead":"This paper claims that pairing the chi-squared feature selection test with support vector machine or random forest improves lung cancer detection, but it does not present the experiments that would support that claim. The manuscript mixes several unrelated datasets and methods, and its only results table appears to come from a different disease classification study.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Central claim is unsupported: the paper reports no results for the claimed lung-cancer experiment; the only results table (Table 3) concerns Alzheimer's disease with different feature-selection methods, so the conclusion about Chi-squared + SVM/RF has no empirical basis.","rationale":"The reader's REJECT verdict is appropriate. My stress test focused on the necessary condition for the central claim: a reported experiment comparing SVM/RF with and without Chi-squared selection on lung cancer data. The manuscript never provides this. Instead, it contains three incompatible dataset descriptions (Section 1.3: 1000 samples/25 features; Section 2: 96 samples/7,129 microarray features; Section 3: TCGA images) and a single results table (Table 3) for Alzheimer's disease using different feature selection methods (MRMR, CFS, SDS) and classifiers (decision tree, Naive Bayes, neural network). The conclusion's attribution of improved accuracy to Chi-squared + SVM/RF is therefore not just weakly supported but entirely unsubstantiated within the paper's own text. This goes beyond transferability concerns raised by the reader: the relevant experiment is absent. A minimal falsifiable check—searching for any reported metric from the described lung cancer experiment—settles the matter. Because the central claim fails this check, the verdict remains REJECT.","tokens_in":8447,"tokens_out":2804,"duration_ms":26676,"concrete_test":"Perform a full-text search for any numeric performance metric (accuracy, recall, precision, F1, runtime) attributed to the lung cancer dataset described in Section 1.3 with Chi-squared selection and SVM or RF classifiers. If no such metric appears, and Table 3 remains the only results table, the central claim in Section 5 has no supporting experiment and the paper should not be accepted as reporting empirical findings.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim—'the SVM model and the Chi-squared Test feature selection technique significantly improve lung cancer diagnostic accuracy' (Section 5)—requires a reported comparison of SVM/RF with and without Chi-squared feature selection on a lung cancer dataset. No such comparison exists in the manuscript. The described lung-cancer experiment in Section 1.3 (25 features, 1000 samples, 65/35 train/test split) is never accompanied by results. Section 2 switches to a different microarray dataset (7,129 features, 96 samples) and discusses kernel-based and SVM-RFE feature selection, not Chi-squared. Section 3 introduces yet another dataset (TCGA images) with SDS feature selection. The only quantitative results, Table 3 and the Section 4 discussion, report classification accuracy, precision, and recall for Alzheimer's disease (AD) using MRMR, CFS, and SDS with decision tree, Naive Bayes, and neural network classifiers—none of which match the claimed Chi-squared + SVM/RF lung cancer experiment. Thus the conclusion is empirically unsupported; Table 3 cannot serve as evidence because it addresses a different disease, different feature selectors, and different classifiers. The paper even self-describes as a 'review' in the abstract and introduction, further undermining the claim that new experimental findings are presented.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper aims to evaluate the combination of Chi-squared feature selection with Support Vector Machine (SVM) and Random Forest (RF) for early lung cancer detection. It describes an intended experiment on a 25-feature, 1000-sample dataset (Section 1.3) and also discusses microarray and radiomics datasets (Sections 2 and 3), but it reports no quantitative results for any lung cancer experiment. The only results table (Table 3) reports classification metrics for Alzheimer's disease using MRMR, CFS, and SDS feature selection with decision tree, Naive Bayes, and neural network classifiers. The conclusion nonetheless states that 'the SVM model and the Chi-squared Test feature selection technique significantly improve lung cancer diagnostic accuracy' (Section 5). This conclusion is not supported by any experiment presented in the manuscript.","tokens_in":8756,"tokens_out":4373,"duration_ms":45503,"significance":"If substantiated, the combination of Chi-squared feature selection with SVM/RF for lung cancer diagnosis could be practically relevant, potentially reducing runtime and improving accuracy in a clinical screening context. However, the manuscript provides no reproducible code, no full experimental protocol, no quantitative predictions, and no results for its central claim. The only quantitative table is for a different disease and different methods, so the significance of the claimed finding cannot currently be assessed.","major_comments":[{"comment":"The lung cancer experiment described in Section 1.3 (a 25-feature, 1000-sample dataset, 65/35 train/test split, Chi-squared feature selection, followed by SVM and RF classification) is never accompanied by any reported results. No accuracy, recall, precision, F1, or runtime figures appear anywhere for this experiment. Consequently, the Section 5 conclusion that 'the SVM model and the Chi-squared Test feature selection technique significantly improve lung cancer diagnostic accuracy' is empirically unsupported.","section":"Section 1.3 and Section 5"},{"comment":"The only quantitative results table, Table 3, reports classification results for Alzheimer's disease (AD), using MRMR, CFS, and SDS feature selection with decision tree, Naive Bayes, and neural network classifiers. This table does not involve lung cancer, Chi-squared selection, SVM, or Random Forest, and therefore cannot serve as evidence for the paper's central claim. Section 4's discussion consists of broad bullet points without any numeric comparison of the described lung cancer models.","section":"Table 3 and Section 4"},{"comment":"The paper self-identifies as a 'review' in the Abstract and Section 1, yet it also describes original experimental comparisons and draws new empirical conclusions. These two modes of presentation have different evidentiary requirements, and the manuscript satisfies neither: a review should synthesize published results with proper citations (which is not done, see reference issues), while an experimental study must provide a complete methods-and-results account, which is absent.","section":"Abstract and Section 1"},{"comment":"Sections 2 and 3 introduce two additional datasets and feature-selection methods: a microarray dataset with 7,129 features and kernel-based/SVM-RFE selection, and a TCGA radiomics dataset with SDS selection. Although performance criteria are listed in Section 2.4, no results are reported for either dataset. The reader cannot determine whether these sections describe the authors' own experiments or prior work, as no quantitative outcomes are provided and the cited references do not supply the missing data.","section":"Sections 2 and 3"},{"comment":"The phrase 'significantly improve' in Section 5 is used without any statistical analysis. There are no significance tests, confidence intervals, or repeated-run variability estimates reported for any model, so the claim of significant improvement is not justified even if the missing experimental results were supplied.","section":"Section 5"}],"minor_comments":[{"comment":"The text refers to 'Backup vector machine (SVM)'; this should be 'Support vector machine.' The same section uses 'superpage' where 'hyperplane' is intended.","section":"Section 2.3"},{"comment":"The name 'Gayon' is a typo; the correct name is 'Guyon' for the SVM-RFE method.","section":"Section 2.2"},{"comment":"The sentence says 'three classification algorithms, random forest and support vector machine' but then lists only two algorithms; 'three' is inconsistent.","section":"Section 1.4"},{"comment":"Figure 1 is captioned as a flowchart of the current work, but the figure actually depicts the SVM maximum-margin hyperplane; no flowchart of the proposed methodology is shown anywhere.","section":"Figure 1"},{"comment":"The confusion matrix in Table 1 is unclear: the row and column labels appear to be interchanged, and the typeset table does not match the standard TP/TN/FP/FN convention described in the bullet points below it.","section":"Table 1"},{"comment":"The section title 'Random release (SDS)' should read 'Stochastic Diffusion Search (SDS),' as the acronym and the algorithm description indicate.","section":"Section 3"}],"recommendation":"reject","confidential_remarks":"The manuscript is not yet a research paper in its current form: the central empirical claim is not reported, the only results table concerns a different disease and different methods, and the reference list contains many entries that are never cited and are unrelated to lung cancer. This may warrant desk rejection rather than a full revision cycle, though the final decision is at the editor's discretion."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe short version: this paper has no reported experiment for its central claim, and the only results table in it is about Alzheimer's disease. I'd send it back.\n\nWhat it does well: it correctly points out that chi-squared feature selection is a standard, cheap way to reduce dimensionality before SVM/RF, and the background descriptions of SVM, RFE, and SDS are broadly accurate, if textbook. The stated aim—comparing SVM and RF with and without chi-squared selection on a lung cancer dataset—is a reasonable and modest question. If the authors had actually run that comparison and reported numbers, the paper would be a small but usable incremental contribution, though not a novel one.\n\nThe soft spots are severe. Section 1.3 describes a 25-feature, 1000-sample dataset with a 65/35 split and says chi-squared selection and SVM/RF were applied. No accuracy, recall, precision, or runtime results for that experiment appear anywhere. Section 2 switches to a 7,129-feature microarray dataset and discusses kernel-based selection and SVM-RFE; Section 3 uses TCGA images with SDS. The only quantitative results, Table 3, report classification metrics for Alzheimer's disease using MRMR, CFS, and SDS with decision tree, Naive Bayes, and neural networks. None of those methods or the disease match the paper's stated experiment. The conclusion that 'the SVM model and the Chi-squared Test feature selection technique significantly improve lung cancer diagnostic accuracy' is therefore unsupported by anything in the manuscript. The stress-test note is right: Table 3 cannot serve as evidence because it is a different disease, different selectors, and different classifiers.\n\nThere is also a citation-hygiene problem. A large fraction of the references are self-citations by the corresponding author on unrelated topics (resource allocation, fraud detection, routing), which look like padding. The paper self-describes as both a review and an experimental study, and it is neither coherently.\n\nWho is this for? No one, as it stands. A reader wanting a compact reminder of chi-squared/SVM concepts might skim the theory, but the lack of results and the mismatched table make it unreliable. This does not deserve a serious referee; it should be desk rejected with clear feedback: run the described experiment, report the numbers, and remove the irrelevant material.","headline":"Central claim is never tested: no results for the described lung-cancer experiment, and the only results table concerns Alzheimer's disease.","tokens_in":9238,"tokens_out":1958,"would_cite":false,"duration_ms":19728,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"SVM with Chi-squared feature selection is claimed to improve lung-cancer diagnosis accuracy and runtime.","keywords":["lung cancer","machine learning","feature selection","Chi-squared test","Support Vector Machine","Random Forest","early detection"],"falsifier":"Run the described lung-cancer experiment—25 features, 1000 samples, three classes, 65/35 train/test split—with and without Chi-squared selection, training SVM and Random Forest; if SVM with selected features does not beat SVM without selection in accuracy or recall across repeated splits, the central claim collapses. The displayed Table 3 (labeled AD) does not involve Chi-squared at all, so it cannot serve as the needed evidence.","tokens_in":8289,"feed_emoji":"🫁","tokens_out":8096,"duration_ms":80946,"temperature":0.7,"pith_summary":"This paper argues that combining the Chi-squared test for feature selection with Support Vector Machine (SVM) classification improves the accuracy, recall, and runtime of lung-cancer detection, and that Random Forest (RF) is a strong comparator especially on complex data. The intended payoff is practical: a cheap statistical filter on features could make automated lung-cancer screening faster and more reliable, which matters because early diagnosis greatly improves survival. The paper also reviews kernel-based selection and SVM-RFE as complementary feature-selection routes for high-dimensional gene-expression data. In short, the authors aim to establish that a simple statistical preprocessing step plus a well-chosen classifier can deliver timely, accurate diagnosis.","feed_headline":"SVM plus Chi-squared selection improves lung cancer diagnosis","feed_subtitle":"A statistical pre-filter on features could make automated screening faster and more accurate.","key_machinery":"The central mechanism is the Chi-squared feature-selection test, a univariate statistical test that measures the dependence between each feature and the target label: for feature $X$ and class $Y$, it computes $\\chi^2 = \\sum \\frac{(O_i - E_i)^2}{E_i}$ over the contingency table, and features with low dependence (low $\\chi^2$) are discarded before classification. This dimensionality reduction is what the paper claims both speeds up SVM/RF training and improves accuracy by removing noise. The SVM classifier, using a maximum-margin hyperplane (with kernel functions such as the Gaussian kernel $\\exp(-\\|X_i-X_j\\|^2/2\\sigma^2)$), and Random Forest, an ensemble of decision trees, are the two classification machines the feature selection feeds.","core_discovery":"The central claim is that applying the Chi-squared test to score and select features before training an SVM classifier yields better lung-cancer diagnosis than training on the full feature set, in both accuracy/recall and runtime. The paper further claims that Random Forest handles complex, high-dimensional data well and remains competitive, and that feature-selection methods generally reduce computational complexity while increasing model accuracy in medical data. The intended upshot is that a univariate statistical filter—dropping features that show little dependence on the target—can be added to existing classifiers as a low-cost step toward earlier screening.","pith_inferences":["The real test of the paper's claim is a direct comparison of SVM with and without Chi-squared on the same lung-cancer data; since Table 3 is about Alzheimer's with different selectors, the lung-cancer evidence remains to be produced.","If Chi-squared selection is genuinely effective, it should generalize to other high-dimensional medical datasets, offering a cheap baseline before deep-learning models are tried.","The runtime-reduction claim depends on actual measured wall-clock times, which the paper does not report; counting feature-selection overhead would clarify the practical gain.","The paper's limited model set leaves open whether tuned neural networks or gradient boosting would outperform SVM once the same filter is applied."],"forward_implications":["If the pairing works, adding Chi-squared selection before SVM yields higher detection accuracy and lower false negatives in lung-cancer screening.","Feature selection reduces runtime, making near-real-time screening feasible on high-dimensional medical data.","Random Forest's ensemble structure gives a robust alternative for complex datasets, as the paper claims.","The combination could become a template for other cancer diagnostics where univariate filters are cheap to apply.","The paper suggests replacing slower genetic-algorithm-based feature selection with statistical tests for faster model development."],"supporting_citations":[{"why":"Baseline review of machine-learning approaches for lung-cancer detection that frames the paper's motivation and comparison.","marker":"[1]"},{"why":"Prior application of the Chi-squared test to lung-cancer prediction; the source of the paper's chosen feature-selection method.","marker":"[3]"},{"why":"Multi-stage SVM lung-cancer detection work cited to justify SVM as a strong classifier.","marker":"[7]"},{"why":"Use of Random Forest and Naive Bayes for clinical disease prediction, grounding the Random Forest comparator.","marker":"[9]"},{"why":"Early lung-cancer detection from image-derived features, supporting the feature-based diagnostic approach.","marker":"[11]"},{"why":"Two-stage SVM-RFE gene-selection strategy, cited as an advanced feature-selection alternative.","marker":"[21]"},{"why":"Multiple SVM-RFE for gene selection in cancer classification, supporting the SVM-RFE discussion.","marker":"[24]"},{"why":"Seminal gene-selection-for-SVM work that underpins the SVM-RFE ranking method.","marker":"[29]"}],"fun_headline_variants":["Chi-squared filter boosts SVM lung cancer detection","Feature selection speeds up and sharpens lung cancer diagnosis","SVM with Chi-squared features beats full-set lung cancer screening","Statistical pre-filter improves machine learning for lung cancer","Reduced features, faster and more accurate lung cancer detection"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper's conclusion that Chi-squared selection improves lung-cancer diagnosis rests on Table 3, which reports Alzheimer's-disease classification results using MRMR, CFS, and SDS selectors, so the transferability of those results to lung cancer with Chi-squared is assumed, not demonstrated.","fun_headline_variants_meta":{"raw":{"variants":["Chi-squared filter boosts SVM lung cancer detection","Feature selection speeds up and sharpens lung cancer diagnosis","SVM with Chi-squared features beats full-set lung cancer screening","Statistical pre-filter improves machine learning for lung cancer","Reduced features, faster and more accurate lung cancer detection"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000207,"raw_usage":{"total_tokens":1311,"prompt_tokens":767,"completion_tokens":544,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":383,"completion_tokens_details":{"reasoning_tokens":468}},"tokens_in":383,"tokens_out":544,"duration_ms":5716,"temperature":1.0,"reasoning_tokens":468,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-10T18:44:48.572648+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the described lung-cancer experiment—25 features, 1000 samples, three classes, 65/35 train/test split—with and without Chi-squared selection, training SVM and Random Forest; if SVM with selected features does not beat SVM without selection in accuracy or recall across repeated splits, the central claim collapses. The displayed Table 3 (labeled AD) does not involve Chi-squared at all, so it cannot serve as the needed evidence.","supporting_citations":[{"cited_title":"Performance analysis of various machine learning -based approaches for detection and classification of lung cancer in humans,","cited_arxiv_id":null,"evidence_quote":"Baseline review of machine-learning approaches for lung-cancer detection that frames the paper's motivation and comparison."},{"cited_title":"Prediction lung cancer–in machine learning perspective,","cited_arxiv_id":null,"evidence_quote":"Prior application of the Chi-squared test to lung-cancer prediction; the source of the paper's chosen feature-selection method."},{"cited_title":"Multi -stage lung cancer detection and prediction using multi -class svm classifie,","cited_arxiv_id":null,"evidence_quote":"Multi-stage SVM lung-cancer detection work cited to justify SVM as a strong classifier."},{"cited_title":"AI -based smart prediction of clinical disease using random forest classifier and Naive Bayes,","cited_arxiv_id":null,"evidence_quote":"Use of Random Forest and Naive Bayes for clinical disease prediction, grounding the Random Forest comparator."},{"cited_title":"Early lung cancer detection using nucleus segementation based features,","cited_arxiv_id":null,"evidence_quote":"Early lung-cancer detection from image-derived features, supporting the feature-based diagnostic approach."},{"cited_title":"Development of two -stage SVM -RFE gene selection strategy for microarray expression data analysis,","cited_arxiv_id":null,"evidence_quote":"Two-stage SVM-RFE gene-selection strategy, cited as an advanced feature-selection alternative."},{"cited_title":"Multiple SVM -RFE for gene selection in cancer classification with expression data,","cited_arxiv_id":null,"evidence_quote":"Multiple SVM-RFE for gene selection in cancer classification, supporting the SVM-RFE discussion."},{"cited_title":"Gene selection for cancer classification using support vector machines,","cited_arxiv_id":null,"evidence_quote":"Seminal gene-selection-for-SVM work that underpins the SVM-RFE ranking method."}],"review_version":1}