WBCHSE HS Artificial Intelligence & Data Science study plan

Write a five-row comparison of AI versus conventional computing covering rule-based logic versus learned behaviour, determinism, handling of noisy input, and maintenance effort, then explain each row aloud in under two minutes

Build a one-page timeline placing symbolic AI, expert systems, the machine-learning era, and modern deep learning with at least one concrete milestone each

Write and run six short Python scripts that use variables, strings, lists, dictionaries, if-elif-else, and both for and while loops, including one loop with a break condition

Define five functions with parameters, default arguments, and return values, then read and fix three deliberately broken snippets covering indentation and off-by-one errors

Sketch the sense-decide-act loop of an intelligent agent and label where a rule-based agent and a learning agent would differ

Create NumPy arrays from lists and files, then practise reshaping, slicing, boolean masking, and vectorised arithmetic across at least ten exercises

Load a CSV into a pandas DataFrame and perform column selection, row filtering with two combined conditions, groupby aggregation, and creation of a derived column

Diagnose and repair a dataset with mixed types, stray whitespace, and missing entries using astype, str.strip, fillna, and dropna, documenting each fix

Build a two-panel Matplotlib figure with axis labels, a legend, a title, and a colour-coded category, then export it as a PNG at readable resolution

Explain in writing when to use a NumPy array rather than a Python list for numerical work, citing memory and speed

Audit a raw dataset for missing values, duplicate rows, inconsistent category spellings, and outliers, recording row and column counts before and after every change

Apply mean or median imputation, deduplication, type conversion, and outlier capping, writing one sentence of justification for each decision

Group the cleaned data by two categorical variables and produce summary statistics plus a correlation heatmap for the numeric columns

Select and justify the correct chart type for five different questions covering distribution, trend over time, comparison between groups, composition, and relationship between two variables

Produce one misleading version and one honest version of the same bar chart, and state exactly which choice caused the distortion

Compute mean, median, mode, standard deviation, and interquartile range for one variable by hand and with pandas, then state whether the distribution is skewed and why

Solve three probability problems using the addition rule, the multiplication rule for independent events, and one conditional probability calculation

Match four real variables to the distribution that best describes them among normal, uniform, and binomial, and justify each match in one sentence

Classify six given problem statements as supervised regression, supervised classification, or unsupervised clustering, and name the target variable where one exists

Explain the difference between correlation and causation using one concrete example from a dataset you have handled

Split a dataset into training and test sets, train a linear regression with scikit-learn, interpret each coefficient, and report RMSE, MAE, and R-squared

Train a logistic regression classifier and report accuracy, precision, recall, F1 score, and a confusion matrix, explaining which metric matters most for an imbalanced case

Run k-means clustering for several values of k, choose k from an elbow plot, and describe each resulting cluster in plain language

Demonstrate overfitting by training the same model at three complexity levels and comparing training versus test error in a table

Write a short justification of the train-test split ratio and the random seed you used, and state what could go wrong without a fixed seed

Implement a perceptron from scratch in NumPy with an explicit weight update rule, and train it to convergence on a linearly separable two-feature dataset

Build a neural network with at least two hidden layers in Keras or PyTorch, plot training and validation loss per epoch, and identify the epoch where validation loss stops improving

Compare ReLU, sigmoid, and tanh activations on the same task and record differences in epochs to convergence and final accuracy

Explain backpropagation and gradient descent in writing, covering the chain rule, the role of the learning rate, and what happens when the learning rate is too large

List four distinct deep learning applications and name the network architecture type typically used for each

Preprocess at least 100 text documents by lowercasing, tokenising, removing stopwords, and applying stemming or lemmatisation, then build a bag-of-words matrix and report its shape

Train a sentiment classifier on labelled reviews and report test accuracy plus three misclassified examples with a written reason for each failure

Load an image as a NumPy array, inspect its shape and channel count, convert it to grayscale, and apply thresholding and edge detection

Describe a convolutional image classifier as five numbered pipeline stages from convolution through pooling, flattening, and dense output

Compare how a chatbot using keyword matching differs from one using a trained intent classifier, listing two limitations of each

Identify three concrete bias or privacy risks in a model you built and propose one specific mitigation for each, such as rebalancing training data or removing an identifier column

Map the three data-protection principles of consent, purpose limitation, and data minimisation onto your own project, stating where each is satisfied or violated

Assemble an end-to-end mini project covering problem statement, data sourcing, cleaning, modelling, evaluation, and stated limitations

Record a five-minute spoken demo of the project and answer three likely examiner questions about your data and metric choices

Sit one full timed mock paper under exam conditions, score it, and list the four weakest topics with a revision action for each