Ranking Madhya Pradesh's Districts on Maternal and Child Health: What a Neural Network Confirms

Which of Madhya Pradesh's 45 districts are furthest behind on maternal and child health — and can you trust the ranking? This study set out to answer both questions at once. The researchers gathered 32 separate health and welfare indicators for every district, spanning things like infant mortality, birth rate, sex ratio, female literacy, household sanitation, and childhood immunization rates — data drawn from India's official Annual Health Survey. They then used a statistical technique called Principal Component Analysis (PCA) to compress those 32 indicators into a single development score per district, and sorted every district into one of four bands: Developed, Developing, Underdeveloped, or Backward.

Indore came out on top; Tikamgarh came out last. Of the 45 districts, 12 landed in "developed," 11 in "developing," 10 in "underdeveloped," and 12 in "backward" — a roughly even four-way split that the authors read as clear evidence of wide regional disparity within a single state.

To check whether the PCA-based ranking was trustworthy rather than a statistical artefact, the authors fed the same district data into a neural network and asked it to independently predict each district's category. It matched the PCA classification for every single district, in both the training and test samples — a strong cross-check that the four-way grouping reflects a real, learnable pattern in the underlying data rather than noise.

What the researchers did

The study drew on secondary data from the Annual Health Survey (AHS) 2012-13 and its Second Updation Bulletin for Madhya Pradesh — a large government survey covering roughly 2,557 sample units (villages and urban blocks) and 2.28 million people across Madhya Pradesh, one of nine "Empowered Action Group" states that together account for a disproportionate share of India's births, infant deaths, and maternal deaths. From this survey, the authors selected 32 indicators spanning seven broad dimensions of wellbeing: fertility reduction, health status of women and children, women's educational status, hygiene and sanitation, electrification, maternal and childcare status, and family welfare.

Principal Component Analysis was used to reduce those 32 correlated indicators down to six underlying components that together captured 81.7% of the total variation across districts. Each district's score on these six components was averaged and rescaled into a single index running from 0 (worst) to 100 (best), then split into four bands using standard normal-distribution cutoffs.

Two independent methods, one confirmed answer

What sets this study apart from similar district-ranking exercises is the validation step. Rather than relying solely on PCA's classification, the authors also trained a three-layer neural network — 32 inputs (one per indicator), a 10-unit hidden layer, and a 4-category output layer — to independently predict each district's development band from the same raw indicator data. The neural network's classifications matched the PCA-based rankings with 100% accuracy on both the districts it trained on and the ones it hadn't seen before. Two different statistical approaches converging on an identical grouping is a meaningfully stronger claim than either method would support alone.

The rankings: who's ahead, who's behind

At the top of the table: Indore, Bhopal, Gwalior, Balaghat, Jabalpur, Hoshangabad, Betul, Ratlam, Narsimhapur, West Nimar, Vidisha, and Neemuch were classified "developed." At the bottom: Tikamgarh, Sidhi, Sheopur, Umaria, Jhabua, Shahdol, Panna, Datia, Damoh, Chhatarpur, Mandla, and Dindori were classified "backward." The gap between the two ends is stark on individual indicators too — infant mortality ranged from 37 per 1,000 live births in Indore to 85 in Panna, and under-five mortality from 46 to 127, roughly a threefold difference within a single state.

A counterintuitive twist: the sex ratio paradox

Not every indicator improved neatly from "backward" to "developed." Sex ratio at birth was actually worse in developed districts (886 girls per 1,000 boys) than in backward districts (935) — a pattern that runs directly against the study's own working hypothesis that socio-economic indicators should improve monotonically with development level. The likely explanation, consistent with a well-documented pattern in Indian demography, is that sex-selective practices are more accessible — not less — where income, healthcare infrastructure, and diagnostic technology are more developed. Economic development, in other words, does not automatically fix skewed sex ratios, and in some respects appears correlated with worse outcomes on this specific measure.

Why it matters for policy

The paper's authors argue this kind of district-level index does something India's state-level averages cannot: it reveals which specific districts need which specific intervention, rather than letting strong performers mask weak ones in a statewide number. They point out that none of Madhya Pradesh's districts had yet met Millennium Development Goal targets for infant and under-five mortality at the time of the study, and recommend concentrating immunization drives and antenatal care outreach specifically in the districts this index flags as backward or underdeveloped.

On sex ratio, the paper's own policy recommendation points the opposite way. Since sex ratio at birth is worst in developed districts (886 girls per 1,000 boys, against 935 in backward districts), the authors call for strict implementation of the Pre-Conception and Pre-Natal Diagnostic Techniques (Prohibition of Sex Selection) Act — the PC&PNDT Act — specifically where the index shows development is highest, not lowest. It's a useful illustration of why a single composite index shouldn't drive a single uniform policy: different indicators within the same index point to different districts needing different interventions.

Reader Q&A

Q: Why use two different methods (PCA and a neural network) instead of just one?

A: PCA alone can be sensitive to how the underlying indicators are weighted and correlated, so a classification based on it could, in principle, be an artefact of the method rather than a real pattern. By training a neural network independently on the same raw data and checking whether it arrives at the same four-way grouping, the authors get a genuine out-of-sample check on whether the PCA classification reflects something real — and in this case, it does, with 100% agreement.

Q: What does it mean that six components explain "81.7% of variance"?

A: The 32 original health indicators are highly correlated with each other — districts with high infant mortality also tend to have low female literacy, poor sanitation, and so on. PCA finds a smaller number of underlying "components" that capture most of that shared variation. Six components capturing 81.7% means most of the meaningful differences between districts can be summarized in six numbers instead of 32, with only about 18% of the original variation lost.

Q: Is the "backward" vs. "developed" labeling based on hard cutoffs or arbitrary categories?

A: The four bands come from standard normal-distribution percentile cutoffs applied to each district's average PCA score — roughly, districts more than two-thirds of a standard deviation below the mean are "backward," those within two-thirds of a standard deviation above or below the mean split into "underdeveloped" and "developing," and those more than two-thirds of a standard deviation above the mean are "developed." It's a statistical convention rather than an arbitrary label, though the specific cutoff (0.6745 standard deviations, the standard normal quartile point) is a modeling choice like any other.

Q: Why does the sex ratio result matter beyond this one study?

A: It's a reminder that composite development indices can average away important, even contradictory, sub-patterns. A district can score well on income, sanitation, and literacy while performing worse on a specific and serious indicator like sex ratio at birth. Policymakers relying only on an aggregate index could miss that a "developed" district still needs targeted intervention on a specific issue — which is exactly the kind of blind spot the authors are arguing district-level indices should expose, not create.

Q: How current is this data, and does it still reflect Madhya Pradesh today?

A: The underlying data is from the Annual Health Survey 2012-13, and the paper was published in 2019 — so the specific numbers (which district ranks where) describe conditions from over a decade ago, not current conditions. The methodology (PCA plus neural-network validation for district ranking) remains a relevant and reusable approach, but readers should treat the district-by-district rankings as a historical snapshot rather than Madhya Pradesh's current standing.

Data source

The study uses secondary data from the Annual Health Survey (AHS) Factsheet 2012-13 and the AHS Second Updation Bulletin 2012-13 for Madhya Pradesh, published by the Office of the Registrar General & Census Commissioner, India — publicly available government data. The AHS covers all 284 districts across nine "Empowered Action Group" (EAG) states (Bihar, Jharkhand, Madhya Pradesh, Chhattisgarh, Odisha, Rajasthan, Uttar Pradesh, Uttarakhand, and Assam), which together constitute roughly 50% of India's population but a disproportionate 60% of births, 71% of infant deaths, and 72% of maternal deaths nationally. For Madhya Pradesh specifically, the AHS sampled 2,557 units (villages in rural areas, blocks in urban areas), covering about 2.284 million people across an average of 456,800 households.

Indicator selection: 7 sectors, 32 indicators

Following the framework proposed by Ram and Chander Sekhara (2006) and Swain et al. (2010), the study grouped indicators into seven broad sectors of human wellbeing: fertility reduction; health status of women and children; educational status of women; hygiene and sanitation; electrification; maternal and childcare status; and family welfare. Thirty-two specific indicators were drawn from these sectors, including Birth Rate, Death Rate, Maternal Mortality Rate/Ratio, Neo-Natal/Post-Neo-Natal/Infant/Under-Five Mortality Rates, Sex Ratio at Birth and Sex Ratio (0-4), Effective (Female) Literacy, household access to electricity/safe drinking water/toilets/LPG/pucca housing/computer or telephone connectivity, and a range of maternal and child healthcare-utilization measures (antenatal checkups, tetanus injections, IFA supplementation, immunization coverage by vaccine type).

Principal Component Analysis

PCA is a multivariate statistical technique that reduces a set of correlated variables into a smaller number of uncorrelated ("orthogonal") components, each a weighted linear combination of the original variables, ordered so the first component captures the largest share of total variation. Applied to the 32 standardized indicators across 45 districts, six components had eigenvalues at or above 1 (the standard retention threshold) and together explained 81.7% of total variance:

Each district's scores on these six components (d1-d6) were confirmed normally distributed via a Kolmogorov-Smirnov test, then averaged into a single mean score per district and rescaled into a 0-100 index using min-max normalization: I_i = [(d(i) - Min(i)) / (Max(i) - Min(i))] × 100.

Classification thresholds

• Backward: less than (mean − 0.6745σ) → index range [-5.43 to -1.71]
• Underdeveloped: (mean − 0.6745σ) to mean → index range [-1.71 to 0.0]
• Developing: mean to (mean + 0.6745σ) → index range [0.0 to 1.71]
• Developed: more than (mean + 0.6745σ) → index range [more than 1.71]

Full district ranking (all 45 districts)

Neural network architecture and validation

The confirmatory model was a three-layer feedforward Multilayer Perceptron (MLP): 32 input units (one per standardized indicator), one hidden layer of 10 units using a hyperbolic tangent activation function, and an output layer of 4 units (one per development category) using softmax activation with a cross-entropy error function. The network's classification accuracy against the PCA-based categories:

Limitations worth flagging

This is a single-state (Madhya Pradesh), single-time-point (AHS 2012-13 data, published 2019) study using secondary government survey data rather than primary data collection, and the 45-district boundaries reflect Madhya Pradesh's administrative geography at that time (the state has since seen some district reorganization). The neural network's "100% accuracy" is validation against the PCA classification's own categories, not against an independent ground truth of actual district development — meaning the two methods agreeing strongly suggests the grouping is statistically robust, but does not independently prove either method captures "true" development correctly. The test-set cell sizes are also very small (as few as one district in a category), which limits how much confidence the accuracy figures alone can support.