Drawbacks of Supervised Learning

What is Supervised Learning?

Supervised Learning is a machine learning technique in which we map the inputs against some specific output.

Some of the commonly used supervised learning algorithms are:

1. Linear Regression
2. KNN (K-Nearest Neighbors)
3. Logistic Regression
4. SVM (Support Vector Machine)
5. Decision Trees
6. Random Forest

Drawbacks of Supervised Learning Algorithms

• High Computational Time: The processing is very expensive computationally & training a large chunk of datasets requires a lot of time.

• Data Preprocessing: Data Preprocessing is required in order to apply the supervised learning algorithms.

• Prone to Overfitting: The supervised learning algorithms can be easily overfitted if not applied correctly.

• Labeled Data Required: We require properly labeled data (i.e. one output mapped against all inputs) to apply supervised learning.

• Cannot give new information: With unsupervised learning, we can gather new pieces of information that were unknown to us, but it is not possible with supervised learning.

• Limited Output: The output is limited to the labels already in the target feature. With Supervised Learning, we can never get a new output; the output will always be one of the labels from the target column.

• Requires Balanced Dataset: For accurate prediction, we must train supervised learning algorithms on the balanced datasets; otherwise, it can get biased for a label with more occurrences.

• Problem with Big Datasets: To apply supervised learning, we need balanced datasets (i.e., nearly the same number of rows corresponding to each label); balancing & preprocessing big datasets is a more significant challenge than making predictions.

• Limited Performance: The supervised learning algorithms are trained to replicate the training dataset; hence they can never outperform the training data.