A Novel Ensemble Linear Regression Scheme Based on Incorporation of the Linear Model Strength Contributed by Each Data Point Co-ordinate Using a Special Weighted Error Metrics Ensembling Scheme Spanning Error Metric Types and Considered Linear Regression Models – An Example of Construction Accident (Fatal Falls) Prediction
Rahul Konda
Department of Civil Engineering, Osmania University, Hyderabad, Telangana, India.
V. S. S. Kumar
Department of Civil Engineering, Osmania University, Hyderabad, Telangana, India and NITTR Chennai, Ministry of HRD, Government of India, India.
Ramesh Chandra Bagadi
Department of Civil Engineering, Osmania University, Hyderabad, Telangana, India and Ramesh Bagadi Consulting LLC (R042752), Madison, 53715, Wisconsin, USA.
N. Suresh Kumar
Department of Civil Engineering, Osmania University, Hyderabad, Telangana, India.
*Author to whom correspondence should be addressed.
Abstract
This study's goal is to present an innovative ensemble linear regression method that takes into account the contribution of each data point's coordinate to the linear model's strength. In order to achieve this, a computation of Construction Accidents (Fatal Falls) Forecasts modelled in accordance with the suggested manner is provided. For the given data points of interest, we first compute the linear regression model and record its R squared value. By leaving out the co-ordinate itself and using linear regression to model the remaining co-ordinates that the parameter takes on, the method involves calculating the forecasts from a linear regression model for each co-ordinate of the parameter of concern (for a specific independent variable co-ordinate of concern). We also record each R squared value's value. The Linear Regression Model Strength coefficient is then assigned to each independent variable co-ordinate as 1 minus the difference between the R squared value of each linear regression model after excluding the independent variable co-ordinate under consideration and the R squared value of the linear regression model initially obtained by using all the data point co-ordinates. We now consider 9 Error Metric parameters namely, MAPE, MAE, MSE, MAD, RMSE, R squared, Adjusted R squared, p-Value and Bias for each of the thusly computed linear regression models and then Range Normalize these values within an open interval of 0 and 1 and use a special weighted formula in arriving at a Weight Coefficient that spans all the aforementioned Error Metrics. We now use a special Weighted Ensembling Scheme and compute the One Step Forecast value that incorporates the effect of all considered 9 Error Metrics spanning all the aforementioned Linear Models. The Standard Linear Regression based prediction method predicts the Construction Accidents – Fatal Falls for 2018 as 421 and our proposed model predicts the same as 300.7881103 ~ 301 which is much better than the Standard Linear Regression Based Prediction scheme.
Keywords: Sansevieria suffruticossa, Linear regression, aerial,, ensemble linear regression, terrestrial, dimorphic, acropetal,, exodermis, unistratose, scattered, collateral,, cencentric