Pith. sign in

REVIEW 3 major objections 4 minor 32 references

Using a consensus of four unsupervised detectors, the paper shows that Ghana's malaria anomalies concentrate in specific regions and that the districts with the most anomalous months are not the ones with the heaviest burden.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · deepseek-v4-flash

2026-08-01 07:03 UTC pith:SK3P5A3K

load-bearing objection Useful regional malaria anomaly maps, but the burden-vs-frequency claim is unsupported by the documented pipeline and the statistical validation is circular—send it back for serious revision, don't desk-reject. the 3 major comments →

arxiv 2607.21559 v1 pith:SK3P5A3K submitted 2026-07-23 cs.AI cs.CEcs.ETstat.APstat.ML

Unsupervised Consensus-Based Anomaly Detection for Spatiotemporal Malaria Incidence in Ghana

T. Ansah-Narh , Y. Asare Afrane This is my paper
classification cs.AI cs.CEcs.ETstat.APstat.ML
keywords malaria surveillanceanomaly detectionconsensus learningspatiotemporal analysisGhanadisease hotspotsunsupervised machine learningroutine health data
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to establish that 'anomalous' months in Ghana's routine malaria record are a real, separable epidemiological phenomenon, not just a label. Running four unsupervised detectors on ten years of monthly regional case counts, the authors flag any region-month that at least three detectors single out. Those flagged months carry far more cases than normal months (Cohen's d = 3.252) and deviate strongly from each region's usual seasonal pattern (d > 1.2). The paper's headline spatial result is that burden and anomaly frequency point in different directions: Tamale carries the largest case load during anomalous months, while a cluster of Ashanti districts shows the most frequent anomalies. If true, surveillance based only on case counts misses an independent signal about where transmission is unstable.

Core claim

On the paper's own terms, the central discovery is that anomaly burden and anomaly frequency are distinct spatial dimensions of malaria risk in Ghana. Over 2014–2023, 1908 region-month observations were scored by four unsupervised algorithms (isolation forest, local outlier factor, autoencoder, elliptic envelope); agreement by three or more defined an anomaly. Anomalous months formed a statistically distinct population with much higher total cases (Cohen's d = 3.252), larger seasonal residuals (d = 1.383), and larger region-standardised deviations (d = 1.245). Spatially, most recurrent anomalies concentrated in Ashanti and Northern regions, with Tamale, Kumasi and Accra as persistent distric

What carries the argument

The load-bearing object is a consensus anomaly score, S, equal to the number of four detectors—Isolation Forest, Local Outlier Factor, autoencoder, and Elliptic Envelope—that classify a region-month as anomalous; S ≥ 3 is called an anomaly. Each detector works on nine engineered features: total cases, previous-month cases, seasonal residual, region z-score, sine/cosine month encoding, trend year, and under-5/over-5 counts. The consensus vote avoids calibrating heterogeneous scores and gives an interpretable confidence level. For district-level maps, the anomaly months assigned to a region are inherited by all districts in that region; district burden during those months is then interpolated

Load-bearing premise

The load-bearing premise is that district-level anomaly frequency can be read from regional anomaly classifications: in the district analyses, every district in a region is assigned the same anomalous months as its region, so the claim that Tamale has high burden while Ashanti districts have high anomaly frequency is only demonstrated at the regional level.

What would settle it

Re-run the pipeline at district level: build the same features from each district's own time series, compute consensus anomaly flags per district, and then compare Tamale's anomaly-associated burden against Ashanti district anomaly rates. If the burden-frequency separation no longer appears—for example, if high-frequency districts are also the high-burden districts—the paper's key spatial claim is refuted.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Malaria burden alone is an incomplete picture: surveillance dashboards that add consensus anomaly flags gain a second, partially independent risk dimension.
  • Persistent anomaly hotspots—Tamale, Kumasi, Accra—are stable enough across 2014–2023 to be treated as priority sites for investigation and resource planning.
  • Because anomalous months are statistically separable with very large effect sizes, anomaly flags can be used as review triggers even in regions with low absolute case counts.
  • The unsupervised consensus framework requires no outbreak labels and can be applied to other endemic diseases with routine surveillance panels.
  • The spatial mismatch between burden and anomaly frequency implies that intervention targeting should consider transmission instability, not just caseload.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • A district-level reanalysis, with anomalies computed per district rather than inherited from the region, is the direct test of the burden-frequency split; if the split disappears, the spatial claim reduces to a regional difference.
  • If the split survives, anomaly frequency could be used as a routine indicator of transmission instability and combined with case counts to produce a two-axis risk classification for districts.
  • Pairing the anomaly flags with rainfall, temperature, intervention and population data—factors the paper names but does not model—would let health authorities test whether the seasonal clustering has environmental drivers.
  • The highlights list includes a stray sentence about 'insurance data'; it matches no analysis in the paper and should be set aside as an editorial artifact.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 4 minor

Summary. The manuscript develops a consensus-based unsupervised anomaly detection framework for monthly malaria surveillance data in Ghana (2014-2023). District-level admissions are aggregated to 16 regions; nine features (total cases, lag, seasonal residual, regional z-score, calendar encoding, trend, age-specific counts) are used to run Isolation Forest, Local Outlier Factor, Autoencoder, and Elliptic Envelope. Region-month observations flagged by at least three of the four detectors are classified as anomalies. The paper reports that anomalies concentrate in Ashanti and Northern Regions, that Tamale carries the largest anomaly burden, and that high anomaly rates cluster in a set of Ashanti districts, leading to a claimed spatial distinction between burden and anomaly frequency. It also reports large and highly significant differences between anomalous and normal months on total cases, seasonal residual, and regional z-score.

Significance. If the findings were robust, the paper would offer a practical, interpretable anomaly-detection workflow for routine malaria surveillance and a descriptive atlas of where Ghanaian transmission departs from expected seasonality. The writing is clear, the four-detector consensus design is sensible, and the authors are explicit about many limitations, including the retrospective nature of the analysis and the lack of causal attribution. However, the two principal claims are not supported by the evidence as presented: the district-level burden-frequency distinction relies on a district-level anomaly map that cannot be derived from the documented region-level pipeline, and the statistical separation in Table 2 is an artefact of testing the same features that define the labels. As a result, the central public-health message of the paper is not established by the current analysis.

major comments (3)
  1. [Section 3.3, Figs. 11 and 13] The district-level analysis is not supported by the methods. Section 2.1 aggregates districts to regions and Section 2.3 applies the four detectors to region-month observations. Figure 11 confirms that all districts in a given region share the same number of anomalous months (37 for Northern, 46 for Ashanti, 11 for Greater Accra). Under the documented pipeline, district anomaly rates are constant within each region: 30.8%, 38.3%, and 9.2%, respectively. Figure 13, however, shows per-district anomaly rates varying up to about 40% and labels Adansi/Afigya Kwabre as high-rate districts. No separate district-level detection is described in Methods, and the Limitations section explicitly states that detection is performed independently for each region. If a district-level run was performed, it is missing from Section 2.3; if not, Figure 13 cannot be produced from the stated methods. The centr
  2. [Section 3.4, Table 2] The statistical validation is circular. The features tested in Table 2 - total_cases, residual, and region_zscore - are inputs to the four anomaly detectors (Section 2.2, Eqs. (1)-(3)), and the anomalous/normal labels are defined by a consensus threshold on those detectors' outputs (Eq. (17)). Large Mann-Whitney statistics and Cohen's d values (e.g., d=3.252 for total cases) are therefore expected by construction and do not independently demonstrate that anomalous months form a distinct epidemiological population. The problem is compounded by the in-sample nature of the baselines: Eq. (2) uses the full-period seasonal mean and Eq. (3) uses the full-period regional mean and standard deviation, so the comparison is not out-of-sample. Validation against external information (e.g., outbreak records, intervention events, or a hold-out period) or against a simpler univariate threshold baseline
  3. [Section 4, Limitations] The paper's own limitations contradict a key element of the results. Section 4 states that 'anomaly detection is performed independently for each region and therefore does not explicitly model spatial dependence among neighbouring regions.' This is consistent with the Methods, but it makes the district-level anomaly-frequency map in Figure 13 impossible to reconcile with the text unless a separate district-level detection exists. The discussion also claims that the framework 'captures both spatial and temporal variation,' yet the detector uses no spatial features and no district-level anomaly output is documented. This internal inconsistency affects the main conclusion of the manuscript, not a peripheral detail.
minor comments (4)
  1. [Highlights] The bullet point 'Advocates for advanced tools to manage complex insurance data effectively' appears unrelated to malaria surveillance and is likely a template leftover. It should be removed.
  2. [Section 2.4.5] The subsection heading 'RBF Spatial Interpolation for Spatial' appears truncated; the final word is missing.
  3. [Figure 15 caption] The caption uses 'Predicted total cases' for surfaces generated by interpolation. 'Interpolated total cases' would be clearer and avoid implying a forecasting model.
  4. [Data availability] The raw data are not public and no code repository is provided. Given the reliance on hyperparameters (contamination 0.10, LOF k=20, AE architecture, consensus threshold), releasing code would materially improve reproducibility.

Circularity Check

2 steps flagged

Statistical separation and district-level burden–frequency claim reduce to the anomaly-defining inputs and region-level flags.

specific steps
  1. self definitional [Section 2.2 (Eqs. 1–6), Section 2.3 (Eqs. 16–17), Section 3.4 (Table 2)]
    "To enable the detection of multivariate anomalies, a set of features was constructed ... Each region–month observation was described by nine variables derived from the raw surveillance data. ... The outputs of the four anomaly detectors were combined into a consensus anomaly score ... observations were categorised as follows: Strong anomaly, if S=4; Moderate anomaly, if S=3; Normal, if S≤2. ... Total malaria cases exhibited the strongest separation ... Cohen's d value of 3.252 ... residual ... Cohen's d reached 1.383 ... Region z-score ... d of 1.245."

    The same variables used to build the feature matrix—total_cases, residual, region_zscore—are the variables tested in Table 2 after the consensus rule (S≥3) has been applied. IF/LOF/AE/EE were fit on that matrix with 0.10 contamination and/or 90th-percentile thresholds, so the 'anomalous' class is defined by extremity in this exact feature space. Comparing anomalous versus normal months on those features is therefore a restatement of the classification rule rather than an independent test; the Cohen's d values summarize the in-sample separation used to define the groups. Eqs. (2)–(3) also compute seasonal and regional means over all 120 months, so each tested month contributes to its own baseline.

  2. self definitional [Section 2.1, Section 3.3 (Figs. 11–13), Section 4 Limitations]
    "data aggregated from the district level to the regional level to enable a coherent spatiotemporal analysis of anomaly patterns ... all ten districts experienced the same number of anomalous months (37 months) ... All ten districts experienced 46 anomalous months ... The uniform occurrence of 11 anomalous months across all ten districts ... Comparison of Figs. 11 and 13 demonstrates that anomaly burden and anomaly frequency were spatially distinct. ... anomaly detection is performed independently for each region."

    With detection performed only at the region-month level, the 37/46/11 anomalous-month counts in Fig. 11 are regional flags copied to every district. Hence each district's anomaly frequency is constant within its region: Northern ≈30.8%, Ashanti ≈38.3%, Greater Accra ≈9.2%. The 'Ashanti cluster of high anomaly rates' is just Ashanti's higher regional anomaly-month count, and Tamale's 'largest burden during anomalies' is just its case volume during Northern's 37 flagged months. The burden-vs-frequency distinction is thus an artifact of assigning regional anomaly months to districts, not a district-level finding; Fig. 13's within-region variation cannot follow from the documented pipeline.

full rationale

The paper's central quantitative claim that anomalous months are a statistically distinct population is a within-sample comparison of the same engineered features used to define the anomaly labels: the detectors flagged extreme observations in a feature space containing total_cases, residual, and region_zscore, and Table 2 then 'confirms' separation on those very features. This is validation by construction rather than independent evidence. Similarly, the headline spatial claim of burden-frequency dissociation depends on district-level anomaly frequencies, but the methods only document region-level detection and the paper itself states that detection is performed independently for each region. Figure 11 shows every district in a region inheriting the same 37/46/11 anomalous months, so the district-level frequency map in Fig. 13 is either an undocumented separate analysis or an artifact of region-level flags. The paper's self-citation to prior work [7] is background context and not load-bearing. There is no machine-checked or externally held-out validation that would break the circularity. Score 7 reflects that the central statistical and spatial claims reduce to the anomaly-defining inputs and region-level flags, though the anomaly detection pipeline itself is a genuine unsupervised analysis.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The central claim depends on a handful of arbitrary hyperparameters (contamination, consensus threshold, k, autoencoder shape) and on in-sample baselines that include the evaluated month. The most consequential assumption is that regional anomaly labels can be projected onto districts, which silently creates the paper's headline spatial distinction. No new physical or conceptual entities are introduced.

free parameters (5)
  • contamination fraction / percentile thresholds = 0.10 for IF and EE; 90th percentile for LOF and AE
    Each detector flags roughly 10% of observations as anomalous; the threshold is chosen a priori and not tuned against known outbreaks or evaluated for sensitivity.
  • consensus threshold S = S ≥ 3 of 4 detectors
    Set to prioritize high-confidence anomalies; the authors note it is user-configurable but provide no analysis of how results depend on it.
  • LOF neighbor count k = 20
    Chosen without stated justification; local density estimates depend on this value.
  • Autoencoder architecture and training = 9-4-9; Adam; batch 32; 150 epochs; early stopping on 10% validation
    Standard choices; no sensitivity analysis or evidence they are optimal for this dataset.
  • In-sample seasonal mean and regional z-score baselines = Computed over the full 2014-2023 period, including the evaluated month
    The seasonal residual (Eq. 2) and region_zscore (Eq. 3) use the entire decade's mean/SD, so each observation's baseline includes itself; this embeds the anomaly signal into the 'expected' behavior and is a design choice that affects which points are labeled anomalous.
axioms (5)
  • domain assumption Each detector's statistical notion of unusualness (isolation path length, local density, reconstruction error, robust Mahalanobis distance) corresponds to an epidemiologically meaningful transmission anomaly.
    Section 2.3 treats these statistical criteria as surveillance-relevant without ground-truth labels or external outbreak records to verify correspondence.
  • domain assumption The nine engineered features capture all relevant dimensions of abnormal malaria transmission.
    Section 2.2 features are restricted to surveillance counts; environmental drivers, intervention coverage, and spatial context are explicitly excluded in Section 4 limitations.
  • domain assumption DHIMS2 monthly case counts are complete and accurate enough for anomaly detection.
    The data availability section and limitations acknowledge reporting completeness and diagnostic practices may vary; the analysis treats counts as observed without correction.
  • ad hoc to paper Consensus of at least three of four detectors increases reliability.
    Section 2.3.5 asserts this trade-off without simulations, benchmarks against single detectors, or comparison to alternative fusion rules.
  • ad hoc to paper District-level anomaly frequency can be represented by the region-level anomaly classification.
    Section 3.3 and Figures 11/13 apply regional anomaly months to all districts; this forces identical anomaly rates within a region and is the load-bearing premise for the burden-vs-frequency distinction. If district-level anomaly detection were performed independently, the reported spatial contrast could vanish.

pith-pipeline@v1.3.0-alltime-deepseek · 28497 in / 13363 out tokens · 128195 ms · 2026-08-01T07:03:35.156451+00:00 · methodology

0 comments
read the original abstract

A consensus anomaly detection framework was applied to monthly malaria surveillance data from Ghana (2014-2023) to identify atypical transmission patterns. Anomalies were highly structured in space and time. Ashanti and Northern Regions accounted for most recurrent anomalies, with persistent hotspots at Tamale, Kumasi, and Accra. A key finding was the spatial distinction between anomaly burden (cumulative cases during anomalous periods) and anomaly frequency (persistence of unusual behaviour). Tamale had the highest burden during anomalies, whereas the highest anomaly rates clustered in Ashanti districts, showing that high-burden areas are not necessarily those with the most frequent anomalous transmission. Anomalous months formed a statistically distinct group, with much higher case counts (Cohen's $d = 3.252$) and large seasonal deviations ($d > 1.2$) compared with normal months. Malaria burden alone provides an incomplete picture of transmission dynamics. By distinguishing where malaria is most prevalent from where transmission behaves most unusually, this framework can strengthen surveillance, prioritise investigations, and support targeted control strategies.

Figures

Figures reproduced from arXiv: 2607.21559 by T. Ansah-Narh, Y. Asare Afrane.

Figure 1
Figure 1. Figure 1: Spatial distribution of cumulative malaria admissions across the administrative regions of Ghana during 2014–2023. Regional colour intensity represents the cumulative malaria burden aggregated from district-level records, with darker shades indicating higher numbers of reported admissions. Circles denote district reporting locations, and marker size is proportional to the corresponding cumulative malaria b… view at source ↗
Figure 2
Figure 2. Figure 2: Monthly malaria admissions across the 16 administrative regions of Ghana from January 2014 to December 2023. Each panel shows the regional aggregate time series, highlighting substantial spatial heterogeneity in malaria burden, seasonal variability, and long-term temporal dynamics across regions. 5 [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Distribution of monthly malaria admissions by region during 2014–2023. Boxplots summarise the median, interquartile range, overall spread, and extreme values of regional monthly case counts, illustrating marked differences in malaria burden and variability across Ghana’s administrative regions. The within-region standardised score (region_zscore) was calculated using Eq. (3): region_zscore = x − µr σr , (3… view at source ↗
Figure 4
Figure 4. Figure 4: Correlation circle from principal component analysis of the nine engineered features. The first two principal components explain 46.2% and 15.4% of the total variance, respectively. Features that cluster together are positively correlated, while features positioned opposite each other are negatively correlated. The total_cases, under5, and over5 form a tight cluster, indicating strong collinearity. residua… view at source ↗
Figure 5
Figure 5. Figure 5: Percentage of region–month observations classified as consensus anomalies (S ≥ 3) across the 16 administrative regions of Ghana during 2014–2023. Values represent the proportion of years in which a given calendar month was flagged as anomalous by at least three of the four unsupervised detection algorithms. Warmer colours indicate higher anomaly frequencies. The figure highlights pronounced seasonal concen… view at source ↗
Figure 6
Figure 6. Figure 6: Temporal distribution of consensus anomaly strength across Ghana’s 16 administrative regions from January 2014 to December 2023. Each cell represents a region–month observation classified according to the number of anomaly detection algorithms in agreement. Normal observations correspond to agreement by two or fewer methods (S ≤ 2), moderate anomalies correspond to agreement by three methods (S = 3), and s… view at source ↗
Figure 7
Figure 7. Figure 7: Distribution of anomaly classifications by region during 2014–2023. Stacked bars show the total number of region–month observations classified as normal (S ≤ 2), moderate anomaly (S = 3), or strong anomaly (S = 4) according to the consensus framework. The figure summarises the relative contribution of each region to the overall anomaly burden and illustrates substantial regional differences in the frequenc… view at source ↗
Figure 8
Figure 8. Figure 8: Two-dimensional UMAP embedding of the 1,908 region–month observations constructed from the nine engineered malaria surveillance features. Each point represents a single region–month observation and is coloured according to its consensus anomaly classification: normal (S ≤ 2), moderate anomaly (S = 3), and strong anomaly (S = 4), where S denotes the number of anomaly detection algorithms in agreement. The p… view at source ↗
Figure 9
Figure 9. Figure 9: Radar-chart comparison of the five strongest anomaly events identified by the consensus framework. Each panel contrasts the feature profile of a strong anomaly detected by all four anomaly detection algorithms with the corresponding long-term regional baseline. Variables include total malaria cases, lagged malaria burden, seasonal residual, within-region standardised anomaly score, and seasonal phase. The … view at source ↗
Figure 10
Figure 10. Figure 10: Pairwise Cohen’s Kappa coefficients measuring agreement among the four anomaly detection algorithms: Isolation Forest (IF), Local Outlier Factor (LOF), Autoencoder (AE), and Elliptic Envelope (EE). Kappa values quantify agreement beyond chance, with larger values indicating stronger consistency in anomaly classification. The matrix provides an assessment of methodological concordance and complementarity w… view at source ↗
Figure 11
Figure 11. Figure 11: District-level malaria burden during anomalous months within Northern, Ashanti, and Greater Accra Regions. Bars represent cumulative malaria admissions recorded during months classified as anomalous by at least three anomaly detection methods (S ≥ 3). Labels indicate the total number of anomalous months experienced by each district between 2014 and 2023. Districts are ranked according to cumulative anomal… view at source ↗
Figure 12
Figure 12. Figure 12: Monthly anomaly strength trajectories for the three districts contributing the largest malaria burden during anomalous periods, compared with the corresponding regional average anomaly strength. Anomaly strength ranges from 0 to 4 and represents the number of anomaly detection methods identifying a given district-month observation as anomalous. The figure illustrates the temporal relationship between dist… view at source ↗
Figure 13
Figure 13. Figure 13: Bivariate representation of district-level malaria burden and anomaly frequency across Ghana during 2014–2023. Bubble size is proportional to cumulative malaria admissions, while colour indicates the percentage of months classified as anomalous (S ≥ 3). Selected districts with particularly high anomaly rates are labelled. The figure distinguishes districts characterised by persistently high burden from th… view at source ↗
Figure 14
Figure 14. Figure 14: Seasonal distribution of anomalous months in Tamale District during 2014–2023. Radial bars indicate the number of months within each calendar month that were classified as anomalous by the consensus framework. The plot illustrates the seasonal concentration of anomaly occurrence within the district exhibiting the highest cumulative malaria burden during anomalous periods. 23 [PITH_FULL_IMAGE:figures/full… view at source ↗
Figure 15
Figure 15. Figure 15: Spatial distribution of district-level malaria hotspots during anomalous months in Northern, Ashanti, and Greater Accra Regions for four representative years (2014, 2016, 2019, and 2022). Continuous surfaces were generated using radial basis function interpolation of the district malaria burden. Warmer colours indicate higher predicted malaria burden during anomalous periods. Labelled districts identify t… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

32 extracted references · 9 canonical work pages

  1. [1]

    Agbemafle, C

    E. Agbemafle, C. Kubio, D. Bandoh, M. Odikro, C. Aza- gba, R. Issahaku, S. Sackey, Evaluation of the malaria surveillance system – adaklu district, volta region, ghana, 2019, Public Health in Practice 6 (2023) 100414. doi:10. 1016/j.puhip.2023.100414

    arXiv 2019

  2. [2]

    M. K. Savi, B. Pandey, A. Swain, J. Lim, D. Callo- Concha, G. R. Azondekon, M. Wahjib, C. Borgemeis- ter, Urbanization and malaria have a contextual relation- ship in endemic areas: A temporal and spatial study in ghana, PLOS Global Public Health 4 (2024) e0002871. doi:10.1371/journal.pgph.0002871

    work page doi:10.1371/journal.pgph.0002871 2024

  3. [3]

    P. U. Eze, N. Geard, I. Mueller, I. Chades, Anomaly detection in endemic disease surveillance data using ma- chine learning techniques, Healthcare 11 (2023) 1896. doi:10.3390/healthcare11131896

    work page doi:10.3390/healthcare11131896 2023

  4. [4]

    K. L. Colborn, E. Giorgi, A. J. Monaghan, E. Gudo, B. Candrinho, T. J. Marrufo, J. M. Colborn, Spatio- temporal modelling of weekly malaria incidence in chil- dren under 5 for early epidemic detection in mozam- bique, Scientific Reports 8 (2018). doi:10.1038/ s41598-018-27537-4

    2018

  5. [5]

    Srimokla, W

    O. Srimokla, W. Pan-Ngum, A. Khamsiriwatchara, C. Padungtod, R. Tipmontree, N. Choosri, S. Saralamba, Early warning systems for malaria outbreaks in thailand: an anomaly detection approach, Malaria Journal 23 (2024). doi:10.1186/s12936-024-04837-x

    work page doi:10.1186/s12936-024-04837-x 2024

  6. [6]

    A. S. Hashemi, M. M. Ghazani, M. Ohlsson, J. Björk, D. Dietler, Surveillance of disease outbreaks using un- supervised uni-multivariate anomaly detection of time- series symptoms, in: Digital Health and Informatics Inno- vations for Sustainable Health Care Systems: Proceedings of MIE 2024, SAGE Publications 1 Oliver’s Yard, 55 City Road, London, EC1Y 1SP,...

    2024

  7. [7]

    Ansah-Narh, Y

    T. Ansah-Narh, Y . A. Afrane, J. B. Tandoh, Bayesian in- ference of nonlinear malaria dynamics in ghana via an ensemble markov chain monte carlo sampler, Expert Systems with Applications 312 (2026) 131540. doi:10. 1016/j.eswa.2026.131540

    arXiv 2026

  8. [8]

    Adu-Prah, E

    S. Adu-Prah, E. K. Tetteh, Spatiotemporal analysis of cli- mate variability impacts on malaria prevalence in ghana, Applied Geography 60 (2015) 266–273

    2015

  9. [9]

    de Souza, L

    D. de Souza, L. Kelly-Hope, B. Lawson, M. Wilson, D. Boakye, Environmental factors associated with the dis- tribution of anopheles gambiae s.s in ghana; an important vector of lymphatic filariasis and malaria, PLoS ONE 5 (2010) e9927. doi:10.1371/journal.pone.0009927

    work page doi:10.1371/journal.pone.0009927 2010

  10. [10]

    Awine, K

    T. Awine, K. Malm, C. Bart-Plange, S. P. Silal, Towards malaria control and elimination in ghana: challenges and decision making tools to guide planning, Global health action 10 (2017) 1381471

    2017

  11. [11]

    M. N. Adokiya, Perspectives of health workers on malaria case referral among pregnant women attending antenatal care in savelugu municipality, ghana: A qualitative de- scriptive study, PloS one 20 (2025) e0319567

    2025

  12. [12]

    E. K. Aidoo, F. T. Aboagye, G. E. Agginie, F. A. Botch- way, G. Osei-Adjei, M. Appiah, R. D. Takyi, S. A. Sakyi, L. Amoah, G. Arthur, et al., Malaria elimination in ghana: recommendations for reactive case detection strategy im- plementation in a low endemic area of asutsuare, ghana, Malaria Journal 23 (2024) 5

    2024

  13. [13]

    A. S. Kolekang, Y . Afrane, S. Apanga, D. Zurovac, A. Kwarteng, S. Afari-Asiedu, K. P. Asante, A. Danso- Appiah, Challenges with adherence to the ‘test, treat, and track’malaria case management guideline among pre- scribers in ghana, Malaria Journal 21 (2022) 332

    2022

  14. [14]

    J. N. Fobil, A. Kraemer, C. G. Meyer, J. May, Neigh- borhood urban environmental quality conditions are likely to drive malaria and diarrhea mortality in accra, ghana, Journal of environmental and public health 2011 (2011) 484010

    2011

  15. [15]

    F. T. Liu, K. M. Ting, Z.-H. Zhou, Isolation forest, in: 2008 Eighth IEEE International Conference on Data Min- ing, IEEE, 2008, p. 413–422. doi:10.1109/icdm.2008. 17

    doi:10.1109/icdm.2008 2008

  16. [16]

    M. M. Breunig, H.-P. Kriegel, R. T. Ng, J. Sander, Lof: identifying density-based local outliers, ACM SIG- MOD Record 29 (2000) 93–104. doi:10.1145/335191. 335388

    doi:10.1145/335191 2000

  17. [17]

    Sakurada, T

    M. Sakurada, T. Yairi, Anomaly detection using autoen- coders with nonlinear dimensionality reduction, in: Pro- ceedings of the MLSDA 2014 2nd Workshop on Machine Learning for Sensory Data Analysis, MLSDA′14, ACM, 2014, p. 4–11. doi:10.1145/2689746.2689747. 30

    arXiv 2014

  18. [18]

    P. J. Rousseeuw, K. V . Driessen, A fast algorithm for the minimum covariance determinant estimator, Technomet- rics 41 (1999) 212–223. doi:10.1080/00401706.1999. 10485670

    arXiv 1999

  19. [19]

    Zimek, R

    A. Zimek, R. J. Campello, J. Sander, Ensembles for unsupervised outlier detection: challenges and research questions a position paper, ACM SIGKDD Explorations Newsletter 15 (2014) 11–22. doi:10.1145/2594473. 2594476

    work page doi:10.1145/2594473 2014

  20. [20]

    Kittler, M

    J. Kittler, M. Hatef, R. Duin, J. Matas, On combining classifiers, IEEE Transactions on Pattern Analysis and Machine Intelligence 20 (1998) 226–239. doi:10.1109/ 34.667881

    1998

  21. [21]

    Pedregosa, G

    F. Pedregosa, G. Varoquaux, A. Gramfort, V . Michel, B. Thirion, O. Grisel, M. Blondel, P. Prettenhofer, R. Weiss, V . Dubourg, et al., Scikit-learn: Machine learn- ing in python, the Journal of machine Learning research 12 (2011) 2825–2830

    2011

  22. [22]

    J. M. K. Aheto, Mapping under-five child malaria risk that accounts for environmental and climatic factors to aid malaria preventive and control efforts in ghana: Bayesian geospatial and interactive web-based mapping methods, Malaria Journal 21 (2022). doi:10.1186/ s12936-022-04409-x

    2022

  23. [23]

    S. P. Kigozi, R. N. Kigozi, C. M. Sebuguzi, J. Cano, D. Rutazaana, J. Opigo, T. Bousema, A. Yeka, A. Gasasira, B. Sartorius, R. L. Pullan, Spatial-temporal patterns of malaria incidence in uganda using hmis data from 2015 to 2019, BMC Public Health 20 (2020). doi:10.1186/s12889-020-10007-w

    work page doi:10.1186/s12889-020-10007-w 2015

  24. [24]

    T. V . Oheneba-Dornyo, S. Amuzu, A. Maccagnan, T. Tay- lor, Estimating the impact of temperature and rainfall on malaria incidence in ghana from 2012 to 2017, Envi- ronmental Modeling & Assessment 27 (2022) 473–489. doi:10.1007/s10666-022-09817-6

    work page doi:10.1007/s10666-022-09817-6 2012

  25. [25]

    Asare, L

    E. Asare, L. Amekudzi, Assessing climate driven malaria variability in ghana using a regional scale dynamical model, Climate 5 (2017) 20. doi:10.3390/cli5010020

    work page doi:10.3390/cli5010020 2017

  26. [26]

    McInnes, J

    L. McInnes, J. Healy, J. Melville, UMAP: Uniform Man- ifold Approximation and Projection for Dimension Re- duction, arXiv e-prints (2018) arXiv:1802.03426. doi:10. 48550/arXiv.1802.03426

    Pith/arXiv arXiv doi:10.48550/arxiv.1802.03426 2018

  27. [27]

    U. N. Nakakana, I. A. Mohammed, B. Onankpa, R. M. Jega, N. M. Jiya, A validation of the malaria atlas project maps and development of a new map of malaria transmis- sion in sokoto, nigeria: a cross-sectional study using ge- ographic information systems, Malaria journal 19 (2020) 149

    2020

  28. [28]

    P. W. Gething, D. L. Smith, A. P. Patil, A. J. Tatem, R. W. Snow, S. I. Hay, Climate change and the global malaria recession, Nature 465 (2010) 342–345. doi:10.1038/ nature09098

    2010

  29. [29]

    C. C. Aggarwal, Outlier Analysis, Springer International Publishing, 2017. doi:10.1007/978-3-319-47578-3

    doi:10.1007/978-3-319-47578-3 2017

  30. [30]

    Chandola, A

    V . Chandola, A. Banerjee, V . Kumar, Anomaly detec- tion: A survey, ACM Computing Surveys 41 (2009) 1–58. doi:10.1145/1541880.1541882

    arXiv 2009

  31. [31]

    Asare, B

    K. Asare, B. K. Nyarko, N. A. B. Klutse, T. Ansah- Narh, R. Damoah, H. A. Koffi, Quantifying the in- fluence of remote climate indices on key climate vari- ables in northern ghana: A comprehensive multivariate approach, Earth Systems and Environment 10 (2025) 577–603. doi:10.1007/s41748-025-00618-x

    work page doi:10.1007/s41748-025-00618-x 2025

  32. [32]

    Owusu, P

    K. Owusu, P. R. Waylen, The changing rainy season cli- matology of mid-ghana, Theoretical and applied clima- tology 112 (2013) 419–430. 31

    2013