6  Intro

Data science combines statistics, maths, specialised programs, AI, ML, etc.. Application of specific principles and analytic techniques to extract information from data used in strategic planning, decision making, etc.

• Handling missing values: If the dataset is large, we can just remove rows with missing data. For smaller datasets, we can substitute it with the mean or average, using df.mean(), or df.fillna(mean),..

• Handling outliers: if the data has garbage value or has extreme values you can drop it. Otherwise, you can maybe try different models, normalize data or use algorithms that are less affected by it

• Difference between error terms and residuals: an error term is generally unobservable and a residual is observable and calculable, making it much easier to quantify and visualize. In effect, while an error term represents the way observed data differs from the actual population, a residual represents the way it differs from sample population data. Residuals actually helps us get an accurate estimate of the error.

• RMSE is used to measure the deviation of the residuals, and MSE is used to find how close is the line to the actual data. Other metrics: MAE (mean absolute error), MAPE (mean absolute percentage error). from sklearn.metrics import …….

• Cost vs loss function: loss function refers to the error of one training example, while a cost function calculates the average error across an entire training set

• Confusion matrix is used to describe the performance of the classification model.

  	Actual Positive	Actual Negative
  Predicted Positive	True Positive	False Positive
  Predicted Negative	False Negative	True Negative

  • accuracy: (true positive + true negative) / total observations.. RMSE is a measure of accuracy in regression
  • error rate: (false positive + false negative) / total observations
  • precision: true positive / (true positive + false positive)
  • specificity: true negative / (true negative + false positive)
  • sensitivity/recall: true positive / (true positive + false negative).. helps to identify the misclassified positive predicitions

  Accuracy metric can be reliable metric only if the dataset is class-balanced. F1 score is a ML evaluation metric that assesses the predictive skill of a model by elaborating on its class-wise performance rather than an overall performance as done by accuracy. If accuracy is 100%, then F1 = 1.

  \[ F1 = 2 * \text{precision} * \text{recall} / (\text{precision + recall}) \]

  Cut off/threshold is the probability that the prediction is true. It represents the tradeoff between false positives and false negatives. Normally, the cut-off will be on 0.5 (random) but you can increase it. All predicted outcome with a probability above it will be classified in the first class and the other in the second class.