Model Evaluation

1. 1Count how many examples belong to each class.
2. 2Find the largest class count.
3. 3Divide it by the total row count.
4. 4Treat that number as your majority-class baseline accuracy.
5. 5Be suspicious if your model is close to that baseline.

0 of 5 done

1. 1Open one ML dataset you may use for a project.
2. 2Count the labels in the target column.
3. 3Calculate the majority-class baseline accuracy.
4. 4Write one sentence on whether accuracy alone is enough.
5. 5Save the sentence for your project README or demo notes.

0 of 5 done

1. 1Check your target column and count each class.
2. 2Choose the simplest method that could be used without training a complex model.
3. 3Run the baseline on the same validation split as your model.
4. 4Measure the same metric for both, such as recall, precision, or F1.
5. 5Keep the baseline result in your notes even if your model barely wins.

0 of 5 done

1. 1Open your current ML notebook or create one for your hackathon idea.
2. 2Print the class counts for your training labels.
3. 3Run DummyClassifier with strategy='most_frequent' on your validation split.
4. 4Save the confusion matrix and your key metric for the baseline.
5. 5Write one sentence saying whether your model beats the baseline.

0 of 5 done

1. 1Write down what class means positive for your project.
2. 2Check whether rows mean actual labels or predicted labels.
3. 3Check the printed label order before naming the cells.
4. 4Point to the biggest error cell and explain it in plain language.
5. 5Connect the error to the product risk, such as missed fraud or blocked users.

0 of 5 done

1. 1Open your project notebook or create a small test notebook.
2. 2Set the positive class for your task in one sentence.
3. 3Generate a confusion matrix on your validation data.
4. 4Save it as confusion_matrix.png or copy it as a table.
5. 5Write two lines explaining the biggest mistake your model makes.

0 of 5 done

1. 1Choose precision if a false alarm wastes money, blocks users, or damages trust.
2. 2Choose recall if missing a real case causes harm, risk, or delay.
3. 3Write down the cost of a false positive in one line.
4. 4Write down the cost of a false negative in one line.
5. 5Pick the metric linked to the bigger cost.

0 of 5 done

1. 1Open your project notes or README draft.
2. 2Name your positive class, such as fraud, disease, spam, defect, or at-risk student.
3. 3Write one false positive cost and one false negative cost.
4. 4Choose precision or recall as your key metric.
5. 5Add one sentence explaining why this metric fits your use case.

0 of 5 done

1. 1Open your project dataset and identify the target label.
2. 2Decide whether rows are independent or need a group split.
3. 3Create train and validation sets with a fixed random_state.
4. 4Print class ratios for train and validation.
5. 5Write one line in your README: dataset size, split ratio, stratified or grouped, and random_state.

0 of 5 done