Machine Learning Foundations Project Presentation
0 likes33 views
The document outlines a project aimed at predicting the electrical energy output of a combined cycle power plant using a dataset of 9568 hourly sensor readings. Three models were developed using multiple linear regression, with Model 1 being selected for its superior performance in terms of R², mean square error (MSE), and mean absolute error (MAE). The project involved thorough data analysis and validation processes, leading to the conclusion that Model 1 provided the best predictive accuracy.
Machine Learning Foundations Project Presentation
- 1. Project Presentation - Machine Language Foundations For Product Managers Course A J B – Amit J Bhattacharyya
- 2. Project Objective & Input Data Characteristics Objective is to build a model to predict the electrical energy output of a Combined Cycle Power Plant, which uses a combination of gas turbines, steam turbines, and heat recovery steam generators to generate power. Input Data – A set of 9568 hourly average data points from sensors at the power plant which can be used to build the model. The sensor readings provide data for the following factors – o Temperature (T) o Ambient Pressure (AP) o Relative Humidity (RH) o Exhaust Vacuum (V) o Net hourly electrical energy output (PE) (Prediction Target) A J B
- 3. Decisions – Approach, Dataset Divisions. Model Finalizations (Feature Groupings) The dataset clearly provides very defined independent variables (AT,AP,RH,V) and dependent/output variable (PE) the energy output from the combined cycle power plant so Multiple Linear Regression approach seems more likely to build Model. There are 9568 observations of the 4 features Ambient Temperature (AT), Ambient Pressure(AP), Relative Humidity(RH) and Exhaust Vacuum(V). So straight off reserved the last 20% ~ 1913 rows for final Test Set. The remaining 7655 was divided 5 groups of 1531 rows each. First Model (MD1) is including all features (AT,AP,V,RH), decision for the Second Model (MD2) was taken from a functional angle to leave out exhaust vacuum and to consider mostly the atmospheric impact parameters of temperature, pressure and relative humidity. Third Model (MD3) was where we selected just AT & V features based on some scatter plots done on first slice of data which helped understand better the relationship of each Independent variable to the target variable (PE). Used Excel Data Analysis Regression on the first 3 lots (from 7655 rows) for the bias & coefficients calculation for the 3 models. And the next 2 lots was used as a validation set for evaluation run on 3 models for comparing their R2 . SSE was calculated for the precited values of the target variable from the 3 models on the Validation Set , SST was also calculated. So, SSE & SST was used to calculate R2 for each of the 3 models to see which model fits the closest and is best able to explain the variances. Model MD-1 was found to be the best. And on the Final Dataset we calculated the Mean Square Error (MSE) and Mean Absolute Error (MAE) for all the 3 models and the results there too reinforced our decision on the selection of the model MD1 from the validation set run. MD1 was the one which ended up with the lowest MSE & MAE. A J B
- 4. Training Set Analysis – Bias & Coefficients calculation for Models
- 5. Validation Set Runs and Analysis – Calculation of R2 for Model Selection
- 6. Final Test Set Run and Vindication of Model Chosen – MSE & MAE