Python, statistics, data preparation, machine-learning models, evaluation and practical projects.
What you will cover
Python
statistics
data preparation
machine-learning models
evaluation
practical projects
Topic list from the institute's programme record.
Certification
Certificate terms for this programme are confirmed by the institute at admission. Any ADIVAS course-completion certificate is an institute certificate; it is not a university degree, an externally accredited qualification or a vendor certification (such as AWS or Microsoft) unless the institute states so in writing.
Curriculum 0.1 · Draft: written for ADIVAS and awaiting academic review and approval. Topics and weeks may change before your batch starts.
Your skill trail
You move through 6 stations. Each week builds one skill you can show, not just a topic you have heard.
1
Python foundations
Write and debug small Python programs and work in notebooks.
Week 1 Python start: values, types and input
Week 2 Control flow and functions
Week 3 Lists, dictionaries and files
Week 4 NumPy, notebooks and reproducibility
2
Data and statistics
Explore and summarise data with pandas, charts and statistics.
Week 5 pandas: loading and exploring data
Week 6 Data cleaning with pandas
Week 7 Descriptive statistics and visualisation
Week 8 Probability, sampling and uncertainty
3
Data preparation
Turn raw data into model-ready features without leakage.
Week 9 Features: encoding, scaling and engineering
Week 10 Train, validation and test splits; pipelines
4
Machine-learning models
Train and compare regression, classification, tree and clustering models.
Week 11 Linear regression
Week 12 Classification with logistic regression
Week 13 Evaluation metrics in practice
Week 14 Decision trees and random forests
Week 15 Clustering and dimensionality
5
Evaluation and responsible AI
Evaluate properly and recognise fairness and privacy risks.
Week 16 Cross-validation, tuning and overfitting
Week 17 Responsible AI: fairness, privacy and consent
6
Applied projects
Build and present an end-to-end project.
Week 18 Text and time-series applications
Week 19 Capstone build
Week 20 Capstone report and presentation
★
Predict Late Fee Payment: A Responsible ML Case
Your project, reviewed by a trainer.
The ADIVAS mastery ladder: how each skill is recognised
Explored You have seen the idea and tried a first example.
Practising You can do it with hints and keep improving.
Proficient You complete the weekly practical task correctly on your own.
Proven You apply it in a project reviewed by your trainer.
Skill-level tracking is being added to the learner workspace. Your trainer already reviews weekly practical work.
What you will be able to do
Write clear Python programs using variables, control flow, functions, lists, dictionaries and files.
Load, clean and explore data with NumPy and pandas inside a reproducible notebook.
Interpret descriptive statistics, distributions and charts and avoid common misreadings.
Prepare features: handle missing values, encode categories, scale and split data without leakage.
Train regression, classification, tree-based and clustering models with scikit-learn.
Evaluate models with suitable metrics, cross-validation and error analysis, and explain overfitting.
Identify fairness, privacy and consent risks and document model limits responsibly.
Deliver an end-to-end project with a notebook, a short report and an honest presentation of results.
Who it suits
School Senior school students who enjoy mathematics and want to see how computers learn patterns from data.
College Engineering, science and computer applications students who want a strong practical base in data science and machine learning.
Professional Working professionals and career changers in IT, analytics or operations who want to build and judge ML models.
Prerequisites: Class 12 mathematics level. Prior programming is helpful but weeks 1 to 4 start from the basics. Tools: Python 3 with Jupyter notebooks or Google Colab (free), NumPy, pandas, matplotlib and scikit-learn, Free public data sets and institute-supplied sample data, Git and a text editor or VS Code.
Weekly plan
Python foundations Weeks 1–4
Week 1: Python start: values, types and input
Theory
Installing Python or using Colab and running a notebook cell
Variables, numbers, strings and f-strings
Input, output and simple arithmetic
Reading error messages
Lab
Write a program that reads a student name and three marks and prints total, average and a formatted result card; fix three deliberately broken programs.
Homework
Write a program to convert rupee amounts to a given currency using a rate typed by the user.
Deliverable
A notebook with five small programs and fixed bugs.
Weekly checkpoint
Code-reading questions and a short coding task.
You will be able to
Write clear Python programs using variables, control flow, functions, lists, dictionaries and files.
Week 2: Control flow and functions
Theory
if, elif, else and comparison operators
for and while loops
Functions with parameters and return values
Testing a function with sample inputs
Lab
Write functions to grade marks, check prime numbers and compute electricity bill slabs using example slabs; test each with at least five cases.
Homework
Write a function that validates a mobile number format and test it.
Deliverable
A notebook of tested functions.
Weekly checkpoint
Trace-the-code questions and a function-writing task.
You will be able to
Write clear Python programs using variables, control flow, functions, lists, dictionaries and files.
Week 3: Lists, dictionaries and files
Theory
Lists, tuples, sets and dictionaries
Slicing, sorting and comprehensions
Reading and writing text and CSV files
Handling errors with try and except
Lab
Read a CSV of 200 students, count students per class with a dictionary, write a summary file and handle a missing file gracefully.
Homework
Write code to find the top three marks without using sort.
Deliverable
A script that reads a CSV and writes a summary.
Weekly checkpoint
Data-structure questions and a file-handling task.
You will be able to
Write clear Python programs using variables, control flow, functions, lists, dictionaries and files.
Week 4: NumPy, notebooks and reproducibility
Theory
NumPy arrays, shape and vectorised operations
Random numbers and random seeds
Writing a readable notebook: markdown, headings, order of cells
Version control basics with Git
Lab
Compare a loop with a vectorised NumPy calculation on 1 million numbers, document the notebook and commit it to a local Git repository.
Homework
Explain why a notebook must run from top to bottom.
Deliverable
Clean, ordered notebook with a Git history.
Weekly checkpoint
Notebook-order questions and a vectorisation task.
You will be able to
Write clear Python programs using variables, control flow, functions, lists, dictionaries and files. Load, clean and explore data with NumPy and pandas inside a reproducible notebook.
Load a public data set of 1,000 rows, answer ten questions with pandas and note surprises in the data.
Homework
List five questions you can ask of a data set with pandas.
Deliverable
Exploration notebook with answered questions.
Weekly checkpoint
Code-reading and data-question exercises.
You will be able to
Load, clean and explore data with NumPy and pandas inside a reproducible notebook.
Week 6: Data cleaning with pandas
Theory
Missing values: detect, drop, fill and why
Duplicates and wrong types
String clean-up and date parsing
Outliers: detect and decide, do not just delete
Lab
Clean a messy fee and attendance data set: standardise names, parse dates, treat missing values with documented choices and flag outliers.
Homework
Write a cleaning log listing each change and why.
Deliverable
Cleaned data set and a cleaning log.
Weekly checkpoint
Scenario questions on cleaning decisions and a pandas task.
You will be able to
Load, clean and explore data with NumPy and pandas inside a reproducible notebook.
Week 7: Descriptive statistics and visualisation
Theory
Mean, median, mode, spread and quartiles
Histograms, box plots, bar and scatter charts
Choosing charts and avoiding misleading axes
Correlation versus causation
Lab
Create six charts from the cleaned data set, label them clearly and write one honest sentence under each; compute correlation between two columns and discuss causation.
Homework
Find a misleading chart in a newspaper and explain why.
Deliverable
Statistics and charts notebook.
Weekly checkpoint
Chart interpretation questions.
You will be able to
Interpret descriptive statistics, distributions and charts and avoid common misreadings.
Week 8: Probability, sampling and uncertainty
Theory
Probability basics and conditional probability
Sampling and why samples can mislead
The normal distribution in simple terms
Confidence and what a p-value does and does not say, in outline
Lab
Simulate coin tosses and dice, draw random samples of different sizes from a data set and show how the mean varies; compute a simple confidence interval.
Homework
Explain in two sentences why a survey of 20 people is risky.
Deliverable
Simulation notebook with explanation.
Weekly checkpoint
Probability questions and a simulation task.
You will be able to
Interpret descriptive statistics, distributions and charts and avoid common misreadings.
Data preparation Weeks 9–10
Week 9: Features: encoding, scaling and engineering
Theory
Categorical encoding: one-hot and ordinal
Scaling: standardisation and min-max
Creating features from dates and text length
Feature relevance and leakage warning
Lab
Take a housing-style data set and build features: encode town, scale area and extract month from date; record which features could leak the target.
Homework
Write why an "outcome date" column must not be a feature to predict an outcome.
Deliverable
Feature-ready table and a leakage list.
Weekly checkpoint
Leakage and encoding scenario questions.
You will be able to
Prepare features: handle missing values, encode categories, scale and split data without leakage.
Week 10: Train, validation and test splits; pipelines
Theory
Why split data and the role of each part
Random versus time-based splits
Fitting transformers only on training data
scikit-learn pipelines
Lab
Split a data set, build a pipeline with imputation, scaling and a baseline model and confirm there is no leakage.
Homework
Draw the data flow of your pipeline on paper.
Deliverable
Pipeline notebook and diagram.
Weekly checkpoint
Leakage detection questions and a pipeline task.
You will be able to
Prepare features: handle missing values, encode categories, scale and split data without leakage.
Machine-learning models Weeks 11–15
Week 11: Linear regression
Theory
Fitting a line and meaning of coefficients
Residuals and the error measures MAE and RMSE
Multiple features and interpreting coefficients carefully
Baselines: predict the mean
Lab
Predict monthly electricity use from household features, compare against a mean-baseline, plot residuals and explain two coefficients.
Homework
Write when linear regression is a poor choice.
Deliverable
Regression notebook with baseline comparison.
Weekly checkpoint
Interpretation questions on coefficients and errors.
You will be able to
Train regression, classification, tree-based and clustering models with scikit-learn. Evaluate models with suitable metrics, cross-validation and error analysis, and explain overfitting.
Week 12: Classification with logistic regression
Theory
Classes, probabilities and thresholds
Logistic regression and decision boundary idea
Confusion matrix: TP, FP, FN, TN
Choosing a threshold for a purpose
Lab
Predict whether a customer will pay on time using a small synthetic data set, build a confusion matrix and show how changing the threshold trades precision and recall.
Homework
Explain which error matters more in your example.
Deliverable
Classification notebook with threshold analysis.
Weekly checkpoint
Confusion matrix calculations.
You will be able to
Train regression, classification, tree-based and clustering models with scikit-learn. Evaluate models with suitable metrics, cross-validation and error analysis, and explain overfitting.
Week 13: Evaluation metrics in practice
Theory
Accuracy, precision, recall, F1 and when each matters
Imbalanced data and why accuracy can mislead
ROC and AUC in outline
Error analysis: study the wrong predictions
Lab
Evaluate three models on an imbalanced fraud-style data set, compute metrics by hand for one, plot a precision-recall trade-off and review twenty wrong predictions.
Homework
Write a half-page metric choice memo for a manager.
Deliverable
Evaluation report with error analysis.
Weekly checkpoint
Metric calculations and scenario questions.
You will be able to
Evaluate models with suitable metrics, cross-validation and error analysis, and explain overfitting.
Week 14: Decision trees and random forests
Theory
How a tree splits data and gini idea in outline
Overfitting and tree depth
Random forests and feature importance with cautions
Comparing with logistic regression
Lab
Train a tree and a forest, compare with logistic regression, change tree depth to see overfitting and interpret feature importance carefully.
Homework
Explain why feature importance is not causation.
Deliverable
Model comparison table.
Weekly checkpoint
Overfitting questions and a model-comparison task.
You will be able to
Train regression, classification, tree-based and clustering models with scikit-learn. Evaluate models with suitable metrics, cross-validation and error analysis, and explain overfitting.
Week 15: Clustering and dimensionality
Theory
k-means clustering and choosing k
Distance, scaling and why it matters
PCA for visualisation, in outline
Judging clusters by usefulness not just numbers
Lab
Cluster shop customers by spend and visits, scale the features, test k from 2 to 6 and describe each segment in business words.
Homework
Name three actions a shop could take for each segment.
Deliverable
Segment notebook with business descriptions.
Weekly checkpoint
Clustering interpretation questions.
You will be able to
Train regression, classification, tree-based and clustering models with scikit-learn.
Evaluation and responsible AI Weeks 16–17
Week 16: Cross-validation, tuning and overfitting
Theory
k-fold cross-validation
Hyperparameters and grid search
Learning curves: more data versus a more complex model
Reporting a result with uncertainty
Lab
Use five-fold cross-validation and a small grid search, plot a learning curve and report the final score on the held-out test set once.
Homework
Explain why the test set must be used once.
Deliverable
Tuning notebook and one-page result note.
Weekly checkpoint
Overfitting and CV questions and a tuning task.
You will be able to
Evaluate models with suitable metrics, cross-validation and error analysis, and explain overfitting.
Week 17: Responsible AI: fairness, privacy and consent
Theory
Where bias enters: data, labels, features and use
Proxy features such as pincode or surname
Group-wise error checks
Privacy, consent and data minimisation
Model cards: documenting purpose, data, limits
Lab
Audit a loan-approval style model on synthetic data: compare error rates between two groups, find a proxy feature, remove it and write a model card with limits.
Homework
Write three questions you would ask before deploying a model in a school or bank.
Deliverable
Fairness audit and model card.
Weekly checkpoint
Scenario questions on bias and privacy.
You will be able to
Identify fairness, privacy and consent risks and document model limits responsibly.
Applied projects Weeks 18–20
Week 18: Text and time-series applications
Theory
Bag-of-words features for short text classification
Simple time-series splitting and moving averages
Baselines for forecasts
Limits of small data
Lab
Classify short customer messages as complaint or enquiry with a bag-of-words model and forecast next month's sales with a moving average baseline.
Homework
Compare a naive forecast with your model and say which to use.
Deliverable
Text and forecast notebook.
Weekly checkpoint
Applied scenario questions.
You will be able to
Train regression, classification, tree-based and clustering models with scikit-learn. Evaluate models with suitable metrics, cross-validation and error analysis, and explain overfitting.
Week 19: Capstone build
Theory
Framing the problem and success measure
Data audit and plan
Building the baseline then improving
Documenting assumptions and limits
Lab
Work on the capstone data set: define the question, clean, build a baseline and improve with one idea at a time, keeping an experiment log.
Homework
Peer-review a classmate's experiment log.
Deliverable
Capstone notebook and experiment log.
Weekly checkpoint
Mid-capstone review.
You will be able to
Load, clean and explore data with NumPy and pandas inside a reproducible notebook. Prepare features: handle missing values, encode categories, scale and split data without leakage. Train regression, classification, tree-based and clustering models with scikit-learn. Evaluate models with suitable metrics, cross-validation and error analysis, and explain overfitting. Identify fairness, privacy and consent risks and document model limits responsibly.
Week 20: Capstone report and presentation
Theory
Writing a two-page report
Honest presentation of results and failures
Responding to questions
Reproducibility check and archiving
Lab
Finalise the capstone, present in ten minutes and rerun the notebook from scratch to confirm results.
Homework
Self-assess with the rubric.
Deliverable
Final notebook, report and presentation.
Weekly checkpoint
Capstone assessment with a viva.
You will be able to
Interpret descriptive statistics, distributions and charts and avoid common misreadings. Identify fairness, privacy and consent risks and document model limits responsibly. Deliver an end-to-end project with a notebook, a short report and an honest presentation of results.
Sample lessons
Week 13
Reading a confusion matrix
A classification model predicts classes such as "late" or "on time". A confusion matrix counts how often each prediction was right or wrong: true positives, false positives, false negatives and true negatives. From these counts we compute accuracy, precision (of the cases we flagged, how many were truly positive) and recall (of the truly positive cases, how many we found). Which measure matters depends on the cost of each kind of error.
Actual / Predicted
Predicted positive
Predicted negative
Actually positive
TP 40
FN 20
Actually negative
FP 10
TN 130
Step by step
Predict labels on the test set only.
Count TP, FP, FN and TN.
Compute accuracy, precision and recall from the counts.
Ask which error is more costly in this problem.
Adjust the threshold or model and compare again.
In real life
A hospital screening tool flags patients for a follow-up test. Missing a sick person (false negative) is worse than an extra test (false positive), so recall is watched closely.
Worked example
Problem. A model on 200 test cases gives TP 40, FP 10, FN 20 and TN 130. Find accuracy, precision and recall.
Solution. Accuracy = (40 + 130) / 200 = 85%. Precision = 40 / (40 + 10) = 80%. Recall = 40 / (40 + 20) = 66.7%. The model finds two out of three real positives, so check if that is enough for the use case.
Common mistakes
Quoting accuracy alone on imbalanced data.
Evaluating on the same data used for training.
Mixing up false positives and false negatives.
Try it yourself
Compute precision and recall for TP 30, FP 20, FN 10, TN 140.
Write one example where precision matters more than recall and one where recall matters more.
Week 10
Splitting data without leakage
A model must be judged on data it has never seen. We split the data into training, validation and test parts. Leakage happens when information from the test part or from the future slips into training, for example by scaling using the whole data set or by using a column that is only known after the event. Leakage makes scores look excellent in the notebook and fail in real use.
1All data
2Split first
3Fit on training only
4Validate and tune
5Test once
Step by step
Set a random seed and split the data first.
Fit imputers, scalers and encoders on the training part only.
Apply the fitted transformers to validation and test parts.
Tune using the validation part or cross-validation.
Use the test part once for the final score.
In real life
An institute predicts fee delays. If the column "reminder sent" is included but is only created after a delay is noticed, the model looks perfect but is useless for deciding who to remind.
Worked example
Problem. A student scales all rows to the range 0 to 1 and then splits the data. What is wrong?
Solution. The scaler used the minimum and maximum of the test rows, so test information influenced training. Split first, fit the scaler on the training part only and then transform the other parts.
Common mistakes
Using a column created after the outcome.
Tuning repeatedly on the test set.
Random split for time-ordered data that should be split by date.
Try it yourself
List three columns in a hospital data set that could leak the outcome.
Rewrite a pipeline so the scaler is fitted on training data only.
Try the questions
The same skills are asked at three levels: School (Classes 6–12), College and Professional. Switch the level and the language (English or Tamil), and say how sure you are before you check: knowing what you are unsure about is part of learning.
Classes 6–12: everyday examples, no prior knowledge. Practice only. Nothing here is graded or saved.
Question 1 / 4 · Mean versus median
Five students scored 40, 45, 50, 55 and 100. Which number better shows a typical score?
Projects
Predict Late Fee Payment: A Responsible ML Case
A small Madurai training institute wants to predict which students may pay their fee late so staff can send gentle reminders. Using a provided anonymised data set of 1,500 past records, build, evaluate and document a model, and decide whether and how it should be used.
Tasks
Explore and clean the data and log every decision.
Build a baseline and at least two models with a proper train, validation and test split.
Evaluate with precision, recall and F1 and analyse wrong predictions.
Check error rates across two groups and identify any proxy feature.
Write a model card and a one-page recommendation including when not to use the model.
Cluster table with sizes and profiles, a chart and a one-page action plan.
How it is assessed
Criterion
Weight
Correct preparation and clustering
35%
Interpretation in business terms
35%
Privacy handling
15%
Communication
15%
Stretch challenge: Compare two values of k and justify the choice.
How you are assessed
Each week: ten questions (code reading, data interpretation, metric calculation, one responsible-AI scenario) and a notebook task of 45 minutes on a new data set.
Practical work. Learners clean data, build and evaluate models and write a short explanation in a notebook; reproducibility, correct metrics and honest limits are checked.
What trainers look for
Weight
Correctness and reproducibility
35%
Appropriate evaluation
25%
Responsible-AI handling
20%
Clarity of explanation
20%
Where this can lead
Prepares you for junior data-analyst and machine-learning assistant tasks.
Prepares you for further study in data science, analytics and applied AI.
These describe skills you will build. ADIVAS does not guarantee jobs, placement or salary.
Other programmes in Artificial Intelligence & Data Science
