Class 9 · Year project

An AI proposal for our valley

Pick a real problem in your valley. Prove it's real, build a rule anyone can read, test it honestly — and say who keeps it running.

The big idea

In Class 8 you trained a model and found out where it failed. This year you do the part that comes before any of that, and the part that comes after.

Problem scoping → data acquisition → data exploration → modelling → evaluation → deployment.

You will choose a problem in your own valley, say exactly what data would show it is real, build a rule-based model that anyone can read, test it honestly, and write a proposal that names who keeps it running. CBSE calls this Part D, and asks that it be linked to a Sustainable Development Goal.

Two things make a proposal serious rather than a poster:

Named data features. Not "we would collect data about water" but "hours of supply per day, per mohalla, written down each morning for four weeks".
An honest evaluation. A rule that is right 7 times out of 10 and whose 3 mistakes you can describe is worth more than a rule you never tested.

📱 Scan to watch: Turning a problem into a proposal —

The golden rule: no data about people

Your model works on measurements and counts, never on individuals.

Consent checklist — tick every line before you collect anything:

Nothing we write down identifies a person, a household or a shop.
No photographs of people, and no names in any file name.
If we ask people questions, the slips are anonymous and the answers are only ever reported as totals.
If the thing we are measuring belongs to someone — a shop, an orchard, a mosque yard — we asked, and they said yes.
We wrote down where each number came from, so anyone can check it.
Anything we are unsure about, we ask the teacher before collecting, not after.

Totals

❌ Not allowed:A list of who walks and who does not

Period 1: Problem scoping

Groups of 3–4. Open problems.ods from the refresher lesson and put the group's problems side by side. Choose one — or take one of the three worked examples below.
Fill in the 4Ws problem canvas in a new Calc file, proposal.ods:
Write the problem statement underneath, in this shape: "[Who] need a way to [do what], because [why]."
Name the Sustainable Development Goal your problem serves, and say in one sentence how. The likely ones here: 2 Zero Hunger · 3 Good Health · 4 Quality Education · 6 Clean Water and Sanitation · 11 Sustainable Cities and Communities · 12 Responsible Consumption · 13 Climate Action.
List the stakeholders — everyone the problem or the solution touches. For each, one line: what do they gain, and what could they lose?
Answer the three questions most proposals skip:
- What would count as good enough? ("Right about 4 times in 5" is an answer. "Perfect" is not.)
- What does a mistake cost? Being wrong in one direction usually costs more than the other.
- Who decides? Where does a person stay in charge? (Class 7 Unit 5: human in the loop.)

Totals

❌ Not allowed:A list of who walks and who does not

Period 2: Data acquisition, and data exploration

Name your data features — the specific things you would measure. Four is plenty. For each one, fill a row in proposal.ods:

Data feature Where it comes from How often Who consents What it misses
Draw a system map in LibreOffice Draw: your problem in the middle, the things that affect it around the outside, arrows for what causes what. Where an arrow goes both ways, say so — those loops are usually where the problem lives.
Acquire the data. Whichever route your problem needs:
- Measure it yourselves over a week (bins, taps, buses, rainfall).
- Use published data and cite the page you took it from.
- Ask — anonymous slips, totals only, following the checklist.
- If there is no time to collect real data, your teacher will give you a prepared dataset from datasets/project/. Say clearly in your slides which of these you used. A proposal built on a prepared dataset is not a lesser proposal; a proposal that hides where its numbers came from is.
Data exploration: put the numbers in Calc and chart them. Then write two sentences — one thing the chart shows, and one thing it cannot tell you.

Period 3: Modelling

You are building a rule-based model — you write the rule, so anyone can read it and argue with it. (A learning-based model would find the rule from examples, as Teachable Machine did in Class 8. Say in your slides why a rule is the honest choice here: your dataset is small, and a rule can be checked by a person.)

Write the rule in plain words first: "If ___ and ___, then flag it as ___."

Build it, either way — both are full marks:

In Calc:

=IF(AND(B2<2;C2>=3);"short supply";"ok")

Or in Python (Thonny):

hours = float(input("Hours of supply today: "))
days_short = int(input("Days short this week: "))

if hours < 2 and days_short >= 3:
    print("short supply")
else:
    print("ok")

If your group is comfortable with Unit 5 Lesson 5.5, put the week's numbers in a list and loop over them instead of typing them one at a time.

Hold data back before you test. Keep at least 6 rows the rule has never been tuned on. Tuning a rule until it fits every row you have proves nothing at all.
Run the rule on the held-back rows and fill in the 2×2 table:
Work out the accuracy: right answers ÷ rows tested, as a percentage.
Then the question that matters more than the percentage: which mistake is worse for your stakeholders — a False Positive or a False Negative? A false alarm about a water shortage wastes somebody's morning. A missed shortage leaves a street without water. Write one sentence saying which way your rule should lean, and adjust one threshold to make it lean that way. Retest on the same rows.

Totals

❌ Not allowed:A list of who walks and who does not

Period 4: Deployment, and presenting

Write the deployment plan in proposal.ods — four short answers:
- Who would run this, and on what? (A register in the office? A spreadsheet? A phone?)
- Who maintains it when the seasons change, the bins change, or the thresholds stop fitting?
- Who decides when the rule flags something — and what do they check before acting?
- AI access: who in your area could not use this, and why? No smartphone, no electricity that week, cannot read English, no one free to write the numbers down. A proposal that only works for people who already have everything is not a proposal.
Name one way your model could be biased: whose situation is missing from your data? (Class 8 Unit 4. If you measured only your own mohalla, say so.)

🕌 From our heritage

Public fountains, wells, channels and bridges across the Muslim world were built as waqf — an endowment whose income was tied to the thing forever. The deed did not only pay for the building; it paid for the upkeep, naming who would repair it and from what revenue. Wherever that income was later cut off, the fountains stopped running, however well they had been built. A proposal that says who maintains it is following an old rule. See Muslim Heritage, "Waqf for Sustainable Water Management".

Make four slides in Impress:
- Slide 1 — the problem and the goal. Your one-sentence problem statement, who has the problem, and the SDG it serves.
- Slide 2 — the data. Your data features, where they came from, your chart, and one thing the data cannot tell you.
- Slide 3 — the rule, and how it did. The rule in plain words, the 2×2 table, the accuracy, and which mistake you chose to lean away from.
- Slide 4 — running it for real. Who maintains it, who decides, who is left out, and the one thing you would do next with another month.
Present for 4 minutes, everyone speaking. The class asks each group one question. The two best questions are always "what did it get wrong?" and "who keeps it running?"

A rule that failed and was understood beats a rule that was never tested. Groups reporting 100% will be asked how many rows they held back, and whether the rule was adjusted after seeing them.

No laptop today?

The whole project runs on paper, and the rubric does not change.

The 4Ws canvas, the data-feature table and the system map go in your notebooks.
Collect the data by hand on a tally sheet.
Chart it on squared paper.
Write the rule as one sentence, then test it row by row against the held-back rows, filling in the 2×2 table by hand.
The four slides become a four-panel poster.

Three worked examples

Take one of these if your group cannot settle on a problem, or adapt it to your own village.

A. Water: when does the tap run? (SDG 6)

Why it matters. As of July 2026, 15,64,303 of 19,24,945 rural households in J&K had a tap water connection — and 3,60,642 still did not. A connection is not the same as water arriving, and nobody writes down when it does or does not. (Source at the foot of this page.)

Data features: hours of supply per day (per mohalla, never per household) · time of day it starts · month or season · rainfall that week · number of households on the line, as a total.

A rule to start from: flag a week as short supply if supply was under 2 hours on 3 or more days. What a mistake costs: a false alarm wastes an inspection; a missed week leaves a street carrying water by hand.

B. Waste: is our segregation actually working? (SDG 11 and 12)

Why it matters. The Achan site takes roughly 550 tonnes a day, and over 11 lakh tonnes of old waste sits there. Segregation fails at the last step as often as the first — sorted waste tipped into one truck was sorted for nothing.

Data features: number of wet items found in the dry bin, counted daily · number of dry items in the wet bin · which block or floor · day of the week · whether the bins were labelled that week.

A rule to start from: flag a bin as not segregated if more than 3 wrong items appear on 2 days in a week. What a mistake costs: a false flag annoys a class; a missed one means a month of sorting wasted.

C. Orchards: is this a scab week? (SDG 2 and 12)

Why it matters. Apple scab spreads in a wet spring, and a grower who sprays too late loses fruit while one who sprays constantly wastes money and chemical. SKUAST-K's AI centre works on exactly this, with far better data than you will have — which is worth saying out loud in your presentation.

Data features: daily rainfall in mm · average temperature · how many hours the leaves stayed wet (rain and cloud as a proxy) · days since the last spray.

A rule to start from: flag high risk if rain of 2 mm or more falls on two days running while the temperature stays between 10 °C and 24 °C. These thresholds are a classroom simplification of a real forecasting model — say so in your slides. What a mistake costs: a false alarm costs a spray; a missed week can cost the crop.

Two more ideas, with one warning. School attendance and highway closures both make good projects — but attendance is personal data about children. Totals only ("23 of 32 present on snow days"), never a list of names, and nothing that predicts an individual child. That rule is not ours; it is the DPDP Act's.

Your own idea is welcome and usually better. It needs your teacher's approval, data a group can actually collect in a week, and no data about people.

Rubric (for teacher and self-assessment)

Each group is described against the skill. Groups are never compared with each other, there is no ranking and no "best project", and a low accuracy does not mean low marks — an honest evaluation of a rule that failed is a strong project.

Totals

❌ Not allowed:A list of who walks and who does not

My reflection sheet (each student fills this in alone; private to them and the teacher)

Our problem was: ______________________
The SDG it serves, and why: ______________________
My job in the group was: ______________________
Our four data features were: ______________________
One thing our data could not tell us: ______________________
Our rule, in one sentence: ______________________
We held back ___ rows. Our accuracy on them was ___ .
Our rule's worst mistake was a False ______ (Positive / Negative), and that matters because: ______________________
Someone our proposal would not reach, and why: ______________________
Who would have to maintain this in two years' time: ______________________
One thing I can do now that I could not do in March: ______________________

The big idea

In Class 8 you trained a model and found out where it failed. This year you do the part that comes before any of that, and the part that comes after.

Problem scoping → data acquisition → data exploration → modelling → evaluation → deployment.

Two things make a proposal serious rather than a poster:

Named data features. Not "we would collect data about water" but "hours of supply per day, per mohalla, written down each morning for four weeks".
An honest evaluation. A rule that is right 7 times out of 10 and whose 3 mistakes you can describe is worth more than a rule you never tested.

📱 Scan to watch: Turning a problem into a proposal —

The golden rule: no data about people

Your model works on measurements and counts, never on individuals.

Consent checklist — tick every line before you collect anything:

Nothing we write down identifies a person, a household or a shop.
No photographs of people, and no names in any file name.
If we ask people questions, the slips are anonymous and the answers are only ever reported as totals.
If the thing we are measuring belongs to someone — a shop, an orchard, a mosque yard — we asked, and they said yes.
We wrote down where each number came from, so anyone can check it.
Anything we are unsure about, we ask the teacher before collecting, not after.

Totals

❌ Not allowed:A list of who walks and who does not

Period 1: Problem scoping

Groups of 3–4. Open problems.ods from the refresher lesson and put the group's problems side by side. Choose one — or take one of the three worked examples below.
Fill in the 4Ws problem canvas in a new Calc file, proposal.ods:
Write the problem statement underneath, in this shape: "[Who] need a way to [do what], because [why]."
Name the Sustainable Development Goal your problem serves, and say in one sentence how. The likely ones here: 2 Zero Hunger · 3 Good Health · 4 Quality Education · 6 Clean Water and Sanitation · 11 Sustainable Cities and Communities · 12 Responsible Consumption · 13 Climate Action.
List the stakeholders — everyone the problem or the solution touches. For each, one line: what do they gain, and what could they lose?
Answer the three questions most proposals skip:
- What would count as good enough? ("Right about 4 times in 5" is an answer. "Perfect" is not.)
- What does a mistake cost? Being wrong in one direction usually costs more than the other.
- Who decides? Where does a person stay in charge? (Class 7 Unit 5: human in the loop.)

Totals

❌ Not allowed:A list of who walks and who does not

Period 2: Data acquisition, and data exploration

Name your data features — the specific things you would measure. Four is plenty. For each one, fill a row in proposal.ods:

Data feature Where it comes from How often Who consents What it misses
Draw a system map in LibreOffice Draw: your problem in the middle, the things that affect it around the outside, arrows for what causes what. Where an arrow goes both ways, say so — those loops are usually where the problem lives.
Acquire the data. Whichever route your problem needs:
- Measure it yourselves over a week (bins, taps, buses, rainfall).
- Use published data and cite the page you took it from.
- Ask — anonymous slips, totals only, following the checklist.
- If there is no time to collect real data, your teacher will give you a prepared dataset from datasets/project/. Say clearly in your slides which of these you used. A proposal built on a prepared dataset is not a lesser proposal; a proposal that hides where its numbers came from is.
Data exploration: put the numbers in Calc and chart them. Then write two sentences — one thing the chart shows, and one thing it cannot tell you.

Period 3: Modelling

Write the rule in plain words first: "If ___ and ___, then flag it as ___."

Build it, either way — both are full marks:

In Calc:

=IF(AND(B2<2;C2>=3);"short supply";"ok")

Or in Python (Thonny):

hours = float(input("Hours of supply today: "))
days_short = int(input("Days short this week: "))

if hours < 2 and days_short >= 3:
    print("short supply")
else:
    print("ok")

If your group is comfortable with Unit 5 Lesson 5.5, put the week's numbers in a list and loop over them instead of typing them one at a time.

Hold data back before you test. Keep at least 6 rows the rule has never been tuned on. Tuning a rule until it fits every row you have proves nothing at all.
Run the rule on the held-back rows and fill in the 2×2 table:
Work out the accuracy: right answers ÷ rows tested, as a percentage.
Then the question that matters more than the percentage: which mistake is worse for your stakeholders — a False Positive or a False Negative? A false alarm about a water shortage wastes somebody's morning. A missed shortage leaves a street without water. Write one sentence saying which way your rule should lean, and adjust one threshold to make it lean that way. Retest on the same rows.

Totals

❌ Not allowed:A list of who walks and who does not

Period 4: Deployment, and presenting

Write the deployment plan in proposal.ods — four short answers:
- Who would run this, and on what? (A register in the office? A spreadsheet? A phone?)
- Who maintains it when the seasons change, the bins change, or the thresholds stop fitting?
- Who decides when the rule flags something — and what do they check before acting?
- AI access: who in your area could not use this, and why? No smartphone, no electricity that week, cannot read English, no one free to write the numbers down. A proposal that only works for people who already have everything is not a proposal.
Name one way your model could be biased: whose situation is missing from your data? (Class 8 Unit 4. If you measured only your own mohalla, say so.)

🕌 From our heritage

Public fountains, wells, channels and bridges across the Muslim world were built as waqf — an endowment whose income was tied to the thing forever. The deed did not only pay for the building; it paid for the upkeep, naming who would repair it and from what revenue. Wherever that income was later cut off, the fountains stopped running, however well they had been built. A proposal that says who maintains it is following an old rule. See Muslim Heritage, "Waqf for Sustainable Water Management".

Make four slides in Impress:
- Slide 1 — the problem and the goal. Your one-sentence problem statement, who has the problem, and the SDG it serves.
- Slide 2 — the data. Your data features, where they came from, your chart, and one thing the data cannot tell you.
- Slide 3 — the rule, and how it did. The rule in plain words, the 2×2 table, the accuracy, and which mistake you chose to lean away from.
- Slide 4 — running it for real. Who maintains it, who decides, who is left out, and the one thing you would do next with another month.
Present for 4 minutes, everyone speaking. The class asks each group one question. The two best questions are always "what did it get wrong?" and "who keeps it running?"

A rule that failed and was understood beats a rule that was never tested. Groups reporting 100% will be asked how many rows they held back, and whether the rule was adjusted after seeing them.

No laptop today?

The whole project runs on paper, and the rubric does not change.

The 4Ws canvas, the data-feature table and the system map go in your notebooks.
Collect the data by hand on a tally sheet.
Chart it on squared paper.
Write the rule as one sentence, then test it row by row against the held-back rows, filling in the 2×2 table by hand.
The four slides become a four-panel poster.

Three worked examples

Take one of these if your group cannot settle on a problem, or adapt it to your own village.

A. Water: when does the tap run? (SDG 6)

Data features: hours of supply per day (per mohalla, never per household) · time of day it starts · month or season · rainfall that week · number of households on the line, as a total.

B. Waste: is our segregation actually working? (SDG 11 and 12)

C. Orchards: is this a scab week? (SDG 2 and 12)

Data features: daily rainfall in mm · average temperature · how many hours the leaves stayed wet (rain and cloud as a proxy) · days since the last spray.

Your own idea is welcome and usually better. It needs your teacher's approval, data a group can actually collect in a week, and no data about people.

My reflection sheet (each student fills this in alone; private to them and the teacher)

Our problem was: ______________________
The SDG it serves, and why: ______________________
My job in the group was: ______________________
Our four data features were: ______________________
One thing our data could not tell us: ______________________
Our rule, in one sentence: ______________________
We held back ___ rows. Our accuracy on them was ___ .
Our rule's worst mistake was a False ______ (Positive / Negative), and that matters because: ______________________
Someone our proposal would not reach, and why: ______________________
Who would have to maintain this in two years' time: ______________________
One thing I can do now that I could not do in March: ______________________