Exam AWS Certified Machine Learning - Specialty topic 1 question 4 discussion

answer should be C

go for C

kNN is not specifically designed for time series forecasting. The best choice is C. Use the Amazon SageMaker Linear Learner algorithm on the single time series consisting of the full year of data with a predictor_type of regressor.

Amazon SageMaker Linear Learner (Regressor) Why? The Linear Learner algorithm can be used for time series regression. Using predictor_type=regressor, it learns trends and patterns in historical data and extrapolates future values. Given limited historical data (only 1 year), a simple linear regression model might perform well as a baseline. While deep learning models (like Amazon Forecast) may be more advanced, Linear Learner is easier to implement and train for a prototype.

A. NO - kNN is not forecasting, it is similarities B. NO - RCF is for anomality detection C. YES - Linear Regression good for forecasting D. NO - we don't want to classify

The reason for this choice is that the Linear Learner algorithm is a versatile algorithm that can be used for both regression and classification tasks1. Regression is a type of supervised learning that predicts a continuous numeric value, such as the air quality in parts per million2. The predictor_type parameter specifies whether the algorithm should perform regression or classification3. Since the goal is to forecast a numeric value, the predictor_type should be set to regressor.

A. Managing Kafka on EC2 is not compatible with least effort requirement B. Doable (in 2024) as Glue supports streaming ETL to consumes streams and supports CSV records -> https://docs.aws.amazon.com/glue/latest/dg/add-job-streaming.html C. Managing an EMR cluster imo is no compatible with least effort requirement D. Firehose supports kinesis data stream as source and it can use lambda to convert CSV records into parquet -> https://docs.aws.amazon.com/firehose/latest/dev/record-format-conversion.html I guess this is a bit old question, pre Glue streaming ETL support (2023) -> https://aws.amazon.com/about-aws/whats-new/2023/03/aws-glue-4-0-streaming-etl/ Thus I'll go for D

This blog wrote Japanese. but its said using LinearLearner for air pollution prediction. https://aws.amazon.com/jp/blogs/news/build-a-model-to-predict-the-impact-of-weather-on-urban-air-quality-using-amazon-sagemaker/

The HyperParameter is . Either “binary_classifier” or “multiclass_classifier” or “regressor”., there is no classifier so the answer is C

Ans should be c

a kNN will require a large value of k to avoid overfitting and we only have 1 year's worth of data - kNNs also face a difficult time extrapolating if the air quality series contains a trend If we had assurances there is no trend in the air quality series (no extrapolation), and we had enough data, then kNN should beat a linear model ... I am inclined to go for C just going off of the cue that "only daily data from last year is available"

Random cut forests in timeseries are used for anomaly detection, and not for forecasting. KNN's are classification algorithms. You would use the Linear Learner as a regressor, since forecasting falls into the domain of regression.

KNN isn't for time series predicting, go for A!

Creating a machine learning model to predict air quality To start small, we will follow the second approach, where we will build a model that will predict the NO2 concentration of any given day based on wind speed, wind direction, maximum temperature, pressure values of that day, and the NO2 concentration of the previous day. For this we will use the Linear Learner algorithm provided in Amazon SageMaker, enabling us to quickly build a model with minimal work. Our model will consist of taking all of the variables in our dataset and using them as features of the Linear Learner algorithm available in Amazon SageMaker

Answer should be A. k-Nearest-Neighbors (kNN) algorithm will provide the best results for this use case as it is a good fit for time series data, especially for predicting continuous values. The predictor_type of regressor is also appropriate for this task, as the goal is to forecast a continuous value (air quality in parts per million of contaminants). The other options are also viable, but may not provide as good of results as the kNN algorithm, especially with limited data. using the Amazon SageMaker Linear Learner algorithm with a predictor_type of regressor, may still provide reasonable results, but it assumes a linear relationship between the input features and the target variable (air quality), which may not always hold in practice, especially with complex time series data. In such cases, non-linear models like kNN may perform better. Furthermore, the kNN algorithm can handle irregular patterns in the data, which may be present in the air quality data, and provide more accurate predictions.

Answer is "C" !!!

answer C