Performance Evaluation of Machine Learning and Deep Learning Techniques

Citations Over Time

Abstract

Prediction is the act of forecasting what will happen in the future. The field of prediction is gaining more importance in almost all the fields. Machine learning techniques have been used widely for predictions also in recent time deep learning algorithms gain more importance. In this paper, we will be performing prediction over a dataset using both machine learning and deep learning techniques, and the performance of each method will be identified and compared with each other. We have used the house price dataset, which consists of 80 features, which will help to explore data visualization methods, data splitting, data normalization techniques. We have implemented five regression-based machine learning models including Simple Linear Regression, Random Forest Regression, Ada Boosting Regression, Gradient Boosting Regression, Support Vector Regression were used. Deep learning models, including artificial neural network, multi output regression, regression using Tensorflow-Keras were also used for regression. The study was further extended to compare the performance of the classification models and hence six machine learning models and three deep learning models including logistic regression classifier, decision tree classifier, random forest classifier, Naïve Bayes classifier, k-nearest neighbor classifier, support vector machine classifier, feed forward neural network, recurrent neural network, LSTM recurrent neural networks were used. The models were also fine-tuned and results were also compared using performance metrics. We have split our dataset in to 70:30 ration for training and testing. In regression models random forest algorithms were performing better with MAE score 0.12, MSE score 0.55, RMSE score 0.230 and R2 score of 0.85 and in deep learning Tensorflow-Keras–based regression model was performing well with MAE score 0.12, MSE score 0.54, RMSE score 0.210 and R2-Score of 0.87, while in the other side, the classification model, random forest model, was performing good with accuracy of 89.21%, and in deep learning classification technique, feed forward neural network model, was performing good with accuracy of 89.52%. Other performance metrics including Cohen kappa score, Matthews correlation coefficient, average precision, average recall, and F1 score were also calculated to compare the performance.

Related Papers

• → GBMVis: Visual Analytics for Interpreting Gradient Boosting Machine(2021)6 cited
• → Prediction of Phishing Sites in Network using Naive Bayes compared over Random Forest with improved Accuracy(2023)6 cited
• → On Application of Machine Learning Models for Prediction of Failures in Production Lines(2021)1 cited
• → Gradient Boosting Machine: A Survey(2019)25 cited
• → Performance Analysis of Heart Disease Prediction System using Novel Random Forest Over Naive Bayes Algorithm with an Improved Accuracy Rate(2023)