Nigeria Disease Risk Intelligence Platform

Cholera Hotspot Allocator - By State

Risk scores computed from total cases, case fatality rate, and cases per 100,000 population across all 36 states and FCT

Cases and Deaths by Year

Top 10 States by Case Load

#	State	Total Cases	Total Deaths	Avg CFR (%)	Cases per 100k	Outbreaks	Risk Level
1	Borno	21,069	1,743	1.73%	0.33	137	High
2	Zamfara	16,155	486	1.27%	0.2	48	High
3	Adamawa	8,779	806	1.7%	0.14	88	Medium
4	Kano	8,707	1,058	1.25%	0.21	100	Medium
5	Bauchi	7,751	1,653	1.07%	0.73	84	Medium
6	Yobe	6,455	725	1.92%	0.23	39	Medium
7	Katsina	4,701	466	0.65%	0.24	43	Low
8	Kwara	4,182	245	2.98%	0.2	25	Low
9	Jigawa	3,058	251	5.8%	0.09	21	Medium
10	Kaduna	2,972	202	1.39%	0.04	69	Low
11	Gombe	1,797	644	2.25%	0.19	26	Low
12	Niger	1,685	249	18.67%	0.16	7	Medium
13	Plateau	1,639	60	2.29%	0.06	16	Low
14	Kebbi	1,369	151	1.66%	0.04	24	Low
15	Oyo	1,283	147	5.62%	0.1	21	Low
16	Taraba	1,282	111	4.76%	0.06	13	Low
17	Ebonyi	1,115	20	0.45%	0.02	24	Low
18	Federal Capital Territory	767	30	1.79%	0.03	15	Low
19	Edo	667	124	13.72%	0.08	6	Medium
20	Sokoto	637	147	8.27%	0.19	14	Low
21	Bayelsa	607	14	0.78%	0.01	34	Low
22	Benue	521	115	11.59%	0.06	5	Medium
23	Rivers	466	35	3.36%	0.79	8	Low
24	Osun	318	21	1.73%	0.02	10	Low
25	Delta	206	20	3.33%	0.03	3	Low
26	Nasarawa	201	11	1.46%	0.02	6	Low
27	Lagos	181	16	3.76%	0.01	13	Low
28	Kogi	148	8	0.98%	0.01	8	Low
29	Enugu	132	25	4.2%	0.01	5	Low
30	Ogun	118	3	0.58%	0.01	5	Low
31	Abia	75	30	0.0%	0.0	20	Low
32	Anambra	53	1	0.62%	0.0	7	Low
33	Cross River	41	100	0.0%	0.0	10	Low
34	Akwa Ibom	25	8	10.67%	0.01	3	Medium
35	Ondo	6	0	0.0%	0.0	5	Low
36	Ekiti	3	2	0.0%	0.0	5	Low

Lassa Fever - National Epidemiological Trend

Weekly surveillance data from NCDC situation reports 2020-2025

Weekly Confirmed Cases and Deaths

Year	Suspected Cases	Confirmed Cases	Deaths	Case Fatality Rate (%)
2020	6,834	1,190	159	13.4%
2021	4,637	511	77	15.1%
2022	7,984	1,042	144	13.8%
2023	9,124	1,271	210	16.5%
2024	10,087	1,311	207	15.8%
2025	8,461	981	180	18.3%

AI-Powered Lassa Fever Outbreak Forecast

Best performing model selected automatically from three candidates based on lowest RMSE on held-out test data

Best Model Selected Random Forest

Selection Criterion Lowest RMSE

Train/Test Split 80% / 20%

Forecast Horizon 12 Weeks Ahead

Actual vs Forecasted Confirmed Cases

Model Disclaimer: This forecast is based on historical temporal patterns only. Actual outbreak dynamics are influenced by rainfall, population movement, sanitation access, and intervention coverage. This tool supports and does not replace expert epidemiological judgment.

Model Validation - Comparing Three ML Algorithms

80/20 train-test split applied chronologically. Models evaluated on unseen held-out data using RMSE, MAE and R2 metrics.

Model	RMSE	MAE	R2	Selected
Linear Regression	20.34	15.06	0.0489	-
Polynomial Regression (degree 3)	17.76	12.9	0.275	-
Random Forest (100 estimators)	17.07	9.52	0.3305	Best Model

RMSE Root Mean Square Error - penalises large prediction errors. Lower is better.

MAE Mean Absolute Error - average prediction error in case counts. Lower is better.

R2 Coefficient of Determination - how much variance the model explains. Closer to 1.0 is better.

Best Model - Actual vs Predicted on Test Set

Water Access vs Cholera Cases - Correlation Analysis

World Bank indicator: People using at least basic drinking water services vs annual cholera burden in Nigeria

Water Access (%) vs Cholera Cases Over Time

Public Health Insight: This analysis uses World Bank indicator SH.H2O.BASW.ZS to examine the relationship between safe water access and cholera burden. Where water access stagnates or declines, cholera outbreaks intensify, supporting evidence-based WASH (Water, Sanitation and Hygiene) investment as the primary prevention strategy for Nigeria.

AI-Driven Resource Allocation Engine

Proportional allocation of national health resources based on composite risk scores. Risk score = 60% case burden + 20% CFR + 20% population-adjusted incidence.

Total ORS Packets 500,000

Total Vaccines 200,000

Total Health Workers 5,000

Allocation Method Risk-Proportional

#	State	Risk Score	Risk Level	ORS Packets	Vaccines	Health Workers	Total Cases	Avg CFR (%)
1	Borno	63.6	High	63,638	25,455	636	21,069	1.73%
2	Zamfara	48.6	High	48,629	19,451	486	16,155	1.27%
3	Adamawa	28.5	Medium	28,517	11,406	285	8,779	1.7%
4	Kano	27.4	Medium	27,416	10,966	274	8,707	1.25%
5	Niger	24.9	Medium	24,914	9,965	249	1,685	18.67%
6	Bauchi	24.5	Medium	24,514	9,805	245	7,751	1.07%
7	Yobe	22.3	Medium	22,313	8,925	223	6,455	1.92%
8	Edo	21.9	Medium	21,913	8,765	219	667	13.72%
9	Benue	21.5	Medium	21,512	8,605	215	521	11.59%
10	Jigawa	20.4	Medium	20,412	8,164	204	3,058	5.8%
11	Akwa Ibom	20.1	Medium	20,112	8,044	201	25	10.67%
12	Sokoto	18.4	Low	18,411	7,364	184	637	8.27%
13	Kwara	18.0	Low	18,010	7,204	180	4,182	2.98%
14	Oyo	14.9	Low	14,908	5,963	149	1,283	5.62%
15	Katsina	14.8	Low	14,808	5,923	148	4,701	0.65%
16	Taraba	13.2	Low	13,207	5,283	132	1,282	4.76%
17	Kaduna	11.3	Low	11,306	4,522	113	2,972	1.39%
18	Gombe	9.7	Low	9,705	3,882	97	1,797	2.25%
19	Plateau	9.3	Low	9,305	3,722	93	1,639	2.29%
20	Enugu	8.8	Low	8,805	3,522	88	132	4.2%
21	Rivers	8.4	Low	8,405	3,362	84	466	3.36%
22	Lagos	8.0	Low	8,004	3,201	80	181	3.76%
23	Delta	7.3	Low	7,304	2,921	73	206	3.33%
24	Kebbi	7.2	Low	7,204	2,881	72	1,369	1.66%
25	Federal Capital Territory	5.8	Low	5,803	2,321	58	767	1.79%
26	Osun	4.4	Low	4,402	1,761	44	318	1.73%
27	Ebonyi	4.1	Low	4,102	1,640	41	1,115	0.45%
28	Nasarawa	3.5	Low	3,502	1,400	35	201	1.46%
29	Bayelsa	3.3	Low	3,301	1,320	33	607	0.78%
30	Kogi	2.4	Low	2,401	960	24	148	0.98%
31	Ogun	1.5	Low	1,500	600	15	118	0.58%
32	Anambra	1.4	Low	1,400	560	14	53	0.62%
33	Abia	0.2	Low	200	80	2	75	0.0%
34	Cross River	0.1	Low	100	40	1	41	0.0%
35	Ekiti	0.0	Low	0	0	0	3	0.0%
36	Ondo	0.0	Low	0	0	0	6	0.0%

Top 10 High-Risk Local Government Areas

LGA-level hotspot ranking for targeted intervention and resource allocation

#	State	LGA	Total Cases	Total Deaths
1	Borno	Kukawa	1,102	4
2	Borno	Maiduguri	1,059	36
3	Niger	Minna	457	13
4	Kaduna	Zaria	262	16
5	Benue	Makurdi	240	30
6	Kano	Gaya	216	2
7	Kebbi	Dandi	216	18
8	Jigawa	Bashuri	181	11
9	Rivers	Angoni	170	20
10	Adamawa	Mubi North	143	12

Methodology, Limitations and Future Work

Academic framing of technical decisions, data quality issues, and research directions

1. Why AI for Disease Surveillance?

Traditional disease surveillance in Nigeria relies on manual aggregation of paper-based reports submitted to the Nigeria Centre for Disease Control (NCDC). This introduces reporting delays of 2 to 6 weeks, inhibiting timely response. This platform applies machine learning to automate risk stratification and outbreak forecasting, enabling proactive rather than reactive public health intervention. AI adds value here by processing multi-source heterogeneous data including epidemiological records, population denominators, and WASH indicators into a unified, interpretable risk signal.

2. ML Model Selection Rationale

Three models were evaluated: Linear Regression (baseline), Polynomial Regression (captures seasonal non-linearity), and Random Forest (handles feature interactions without assuming linearity). A chronological 80/20 train-test split was applied to preserve temporal structure. Random shuffling would constitute data leakage in time-series forecasting. The best model was selected objectively by lowest RMSE on the held-out test set. Features used: week index, month as seasonality proxy, and year as trend proxy.

3. Risk Score Methodology

The composite risk score for cholera hotspot classification is computed as:

Risk Score = (Cases / Max Cases) x 60 + min(CFR, 10) x 2 + min(Cases per 100k, 50) x 0.4

This weights absolute case burden most heavily at 60%, while incorporating mortality severity via CFR and population-adjusted incidence to avoid bias toward densely populated states. States scoring 40 and above are High risk, 20 and above are Medium, and below 20 are Low risk.

4. Data Limitations and Acknowledged Gaps

Gap The HDX cholera CSV was retrieved empty at 4 bytes. Cholera analysis relies entirely on the NGA historical outbreak dataset covering 1971 to 2023 with inconsistent LGA-level reporting.

Gap Lassa fever data contains no state-level breakdown, only national weekly aggregates from NCDC situation reports. Sub-national Lassa hotspot mapping is not possible with current data.

Gap The NGA dataset contained LGA names entered in the State column such as Maiduguri, Jere, Mafa, and Ganjuwa. These were corrected programmatically by mapping to their parent states using a validated lookup table.

Gap World Bank water access data is national-level only. State-level WASH data from NWASCO or JMP would significantly improve spatial risk modelling.

Gap The ML forecast uses only temporal features. Incorporating rainfall, temperature, displacement data, and vaccination coverage would substantially improve predictive accuracy.

5. Future Research Directions

Integrate LSTM neural networks for sequence-aware outbreak forecasting

Add state-level choropleth map using Nigeria GeoJSON for spatial risk visualisation

Incorporate climate variables from ERA5 reanalysis data as model features

Connect to NCDC live API for real-time surveillance dashboard updates

Expand resource allocation to include cold chain logistics for vaccine distribution

Apply SHAP values for model interpretability critical for public health policy uptake

6. Data Sources

Dataset	Source	Coverage	Indicator
Lassa Fever Weekly Reports	NCDC Nigeria	2020 to 2025	Suspected, confirmed, deaths
Cholera Outbreak Records	HDX and Academic Literature	1971 to 2023	Cases, deaths, CFR by state
Water Access	World Bank Open Data	2000 to 2025	SH.H2O.BASW.ZS
Population Denominators	World Bank and NBS	1971 to 2023	National population estimates