Why does a Kingston machine learning class care about rare cases?
Because accuracy hides them. In our Kingston project only 15 of 197 Melbourne stations are as busy on Saturday as on a weekday. A plain model scores 93.4% and finds two of them; weighting the rare class finds nine but raises 42 false alarms. Choosing between them is the real skill.
Southland station is unusual. In FY24-25 it averaged more entries on a Saturday than on a normal weekday, something only four of the 197 stations with at least 500 weekday entries managed. Teaching a model to recognise that pattern from weekday travel times is a perfect lesson in imbalanced data, because the easy answer, "never Saturday-busy", is already right 92.4% of the time.
Facts last verified 10 October 2026. Teaching is online; no Kingston branch is claimed.
Each course here gets learners evaluating models honestly, not just training them. The free lesson shows which suits.

Machine learning with Python, including precision, recall and imbalanced data.
See the syllabus
Visual machine learning with Orange before writing code.
See the syllabus
Data science from cleaning to modelling and evaluation.
See the syllabus
A complete machine learning course for study and work.
See the syllabusBrowse the course atlas for more than one hundred options and use the coding roadmap to check prerequisites.
The four we are known for
These run underneath everything above. Every one is live and online, placed by ability rather than by age, and the first class is free.

Python, web and AI projects where the learner still owns the thinking.
See the syllabus
Automate the work you already do, then let AI carry part of it.
See the syllabus
Train a model, read what it learned, and be able to say why it is wrong.
See the syllabus
Run AI coding agents on real work without losing control of the codebase.
See the syllabusTotals for the City of Kingston (ABS LGA23430), read from 2021 Census QuickStats on 10 October 2026. This is the Victorian council, not any other Kingston.
| Item | Kingston |
|---|---|
| Usual residents | 158,129 |
| Median age (years) | 41 |
| Families counted | 43,676 |
| Private dwellings | 67,560 |
| People per household | 2.5 |
| Motor vehicles per dwelling | 1.8 |
| Measure | Kingston | Victoria |
|---|---|---|
| Separate houses | 58.8% (35,855) | 73.4% |
| Flats or apartments | 15.0% (9,155) | 12.1% |
| Bachelor degree or above | 31.1% (40,750) | 29.2% |
| Professionals | 26.5% (21,568) | 25.0% |
9,279 Kingston residents, 5.9%, were in this age band.
There were 8,611 under-fives, or 5.4% of the population.
With Melbourne in lockdown on Census night, 30.0% of Kingston workers (24,383) worked from home; that share reflects the restrictions.
QuickStats percentages are rounded and small counts are randomly adjusted by the bureau to protect privacy. We use each published value as it stands and never add categories up.
From the FY24-25 entries file we kept the 197 stations averaging at least 500 entries on a normal weekday. We labelled a station "Saturday-busy" if its Saturday average reached 80% of its weekday average: 15 do. Each station's features are the shares of its weekday entries in the file's five time bands. We trained logistic regression and tested every station while leaving it out of training.
| Model | Accuracy | Saturday-busy found | False alarms |
|---|---|---|---|
| Always say "no" | 92.4% | 0 of 15 | 0 |
| Logistic regression, plain | 93.4% | 2 of 15 | 0 |
| Logistic regression, balanced weights | 75.6% | 9 of 15 | 42 |
| Station | Saturday-busy? | Plain score | Weighted score |
|---|---|---|---|
| Southland | yes | 0.34 | 0.95 |
| Mordialloc | yes | 0.06 | 0.44 |
| Carrum | yes | 0.04 | 0.31 |
| Moorabbin | no | 0.11 | 0.64 |
| Edithvale | no | 0.11 | 0.63 |
| Chelsea | no | 0.06 | 0.42 |
| Westall | no | 0.06 | 0.41 |
| Parkdale | no | 0.05 | 0.34 |
| Aspendale | no | 0.05 | 0.33 |
| Mentone | no | 0.03 | 0.21 |
| Cheltenham | no | 0.03 | 0.19 |
The plain model beats the do-nothing answer by one point and flags no Kingston station at all. With 182 "no" stations to 15 "yes", the cheapest way to score well is to ignore the rare class.
Balanced weights count each rare example more, here about 6.6 times a common one. Recall rises from 2 of 15 to 9 of 15, and Southland jumps from 0.34 to 0.95.
Weighting also flags 42 stations that are not Saturday-busy, Moorabbin and Edithvale among them, and still misses Carrum and Mordialloc. Only 9 of 51 alarms are right.
If missing a Saturday-busy station is costly, the weighted model is better. If false alarms are costly, the plain one is. On accuracy alone, the simple baseline is almost as good as either. Learners write down the cost of each mistake before choosing.
Weekday time shares are a weak signal for Saturday use, which is why neither model is strong; that limit is part of the lesson. Station entries are departmental estimates rounded to the nearest 50, and which stations are in Kingston is our test against the ABS outline. Bonbeach falls below the 500-entry cut and is left out.
The idea fits in one line: weight each class by the number of samples divided by twice that class's count.
| Class | Stations | Weight per station |
|---|---|---|
| Saturday-busy | 15 | about 6.6 |
| Not Saturday-busy | 182 | about 0.54 |
scikit-learn applies the same formula when class_weight is set to "balanced", n_samples divided by (n_classes times the class count). We wrote it by hand with NumPy so learners can see every step, then compare with the library.
Sort a bag of mostly red counters with a few blue ones, and see why "always red" seems smart.
Train both models, draw the confusion tables, and argue which mistake matters more.
Re-check a work model built on rare events, such as faults or returns, with recall alongside accuracy.
From the 2025 school locations list and the metropolitan entries file.
The 2025 list has 23 primary, 5 secondary and 2 specific-purpose government schools in Kingston.
Twelve stations sit inside the council outline. Cheltenham was the busiest in FY24-25, with 717,500 entries.
Term 1 for Victorian government schools runs from Thursday 28 January to Thursday 25 March 2027.
Council outline
Figures use the ABS boundary of the City of Kingston, about 91.4 square kilometres. School addresses inside it include Cheltenham, Mentone, Mordialloc, Chelsea, Dingley Village and Patterson Lakes.
Rough starting points by age; the free lesson fine-tunes them.
Scratch games, sorting and simple patterns.
Scratch Coding for KidsAI Literacy for Kids (Ages 8 to 14)First Python, tables of numbers and simple predictions.
Python and AI for KidsMaths Through CodingModels, evaluation and imbalanced data.
AI and Machine Learning for TeensOrange Data Mining Course for TeensEnd-to-end projects for study and careers.
Data Science CourseAI and ML MasterclassDeciding what "good" means, which no tool can do for you.
Automated tools usually rank models by accuracy unless told otherwise. On this data that would pick the plain model, 93.4%, which never flags a Kingston station. Nothing in the output warns that the rare class has been ignored.
A learner who has seen recall jump from 2 to 9 while false alarms reach 42 knows to ask which mistake is worse before any model is chosen. That question shapes everything that follows, including how an AI tool is set up.
Kingston learners use AI tools well when they choose the measure that matters. The longer argument is in why teenagers should still learn to code in 2026.
Every class runs on video, for learners in Mentone, Chelsea, Dingley Village or elsewhere in Kingston.
The learner shares their screen and builds while the teacher explains.
Teaching tracks the Victorian term dates and breaks for holidays.
A complete lesson free of charge, with a course suggestion afterwards.
Fifteen to twenty learners at the same level, drawn from many suburbs.
Two fixed sessions weekly, about eight each month.
A 4:45pm Kingston class is 11:15am India time under Melbourne daylight saving and 12:15pm once it ends.
Daylight saving dates
Melbourne observes daylight saving from Sunday 4 October 2026 to Sunday 4 April 2027. India keeps one time all year, so our times are always in Melbourne time.
Fees are simple and identical for every family.
Lesson one is free, with clear advice at the end.
Group option: a month of lessons, usually eight, fifteen to twenty learners together.
Private option: a month of lessons, usually eight, one learner with one teacher.
Kingston families pay in US dollars. You pay nothing until the trial agrees a course and a weekly slot; holiday and missed-lesson rules sit on the pricing page.
Rated 4.9 across 547 Google reviews. These are real reviews, reproduced as written.
★★★★★
"The one step solution for my son. Modern Age Coders make learning coding so simple that kids love it. The teachers explain complex concepts clearly with practical exercises and interactive content."
Ria Mukherjee
Parent
★★★★★
"Modern Age Coders has been a game-changer for me. I struggled to grasp IT concepts and coding before joining, but their classes transformed everything. I can now confidently write complex programs with ease."
Samriddha Mondal
Student
★★★★★
"One of the most wonderful education centres out there. Education is not limited to school syllabus but focuses on skill development."
Vansh Agarwal
Student
★★★★★
"My child Dhairya is really enjoying the Modern Age Coders classes. This is his first online class and he eagerly looks forward to it. I can already see his improvement, and the teachers are very cooperative."
Sonam Oswal
Parent of Dhairya
★★★★★
"Modern Age Coders have wonderful teachers who teach in a clear, easy and practical way. The teacher boosts students' confidence and inspires them to learn without hesitation."
Sonu Goyal
Parent
★★★★★
"I highly recommend this computer coding class! The teachers are incredibly knowledgeable and passionate about coding."
Ritu Kedia
Parent
Tell us the learner's age and interests. A younger learner might start a Scratch game; a teenager might train the Saturday station model.
Questions from Kingston parents and adult learners.
The 2021 Census counted 158,129 residents in the City of Kingston, Victoria, with a median age of 41.
An imbalanced data set has one class much rarer than the other, such as 15 Saturday-busy stations among 197. Models then tend to ignore the rare class.
Class weights make mistakes on some classes count more during training. Balanced weights give each class equal total influence, so rare examples matter as much as common ones.
When one answer is right 92.4% of the time, a model can score well by always giving it. Recall and precision show what happens to the rare cases.
The ideas suit children from about seven through games and sorting. Training real models in Python usually starts around 13.
Yes. Adults can learn data science and machine learning in groups or one to one.
No. Every class is live online; we have no office in Australia, and the phone number we list is Indian.
The first lesson is free. Group classes cost USD 100 a month for two live lessons a week, about eight a month, in groups of fifteen to twenty, and one-to-one classes cost USD 150 a month. You agree the course and time before paying.
Into groups of fifteen to twenty by level and pace, not by age or suburb, with one-to-one lessons for anyone whose timing no group fits.
Cheltenham, with 717,500 entries, among the twelve stations inside the council outline by our calculation.
Try the Melbourne page for the city, the Victoria guide for the state, and coding for kids in Melbourne for younger learners.