mcdonaldabdullah
Student interesting in various topics such as Machine Learning, AI, Big Data, IoT, etc.
Pinned Repositories
Comparison-of-Predictive-Models-for-Rainfall-Prediction-using-Big-Data-Technologies
Rainfall is a form of precipitation and is responsible for providing most of the freshwater for animals and plants. Machine learning can be used to analyze data trends to develop a model. Deep learning on the other hand focuses more on using images specifically to analyze data. Trying to understand the patterns of rainfall to predict it has proven to be a difficult undertaking, as seen by the various research using machine learning and deep learning for this problem. When implementing a solution to this rainfall prediction problem, a vast amount of computational resources are usually required to execute it. Thus arises a need to properly store and analyze the data to effectively approach the prediction aspect. This paper investigated the comparison of predictive models for rainfall prediction using big data technologies and radar rainfall images. The literature on state-of-the-art prediction models was investigated and compared to survey which models could achieve satisfactory prediction results in combination with big data technologies. The models chosen were Random Forest Regressor and Deep LSTM and were used to predict 1,2, and 3 days ahead using monthly rainfall data. Results from this study showed that the Deep LSTM model performed better than the Random Forest Model for sequence lengths of 4, 8, and 12 when predicting 1, 2, and 3 months ahead.
Daily-Rainfall-Prediction-Using-Radar
Reliable daily rainfall predictions can play an important role in a) watershed management, b) disaster management, and c) helping people to plan their day. Numerous factors can have an effect on the patterns of rainfall and therefore it can be difficult to predict. Recent papers studied deep learning for rainfall prediction using various prediction models with an emphasis on short-term predictions (hourly), however, few have looked at daily rainfall prediction using deep learning. This paper discusses a deep learning model, ConvLSTM, for daily rainfall prediction using various sequence lengths of radar images and predicting 1, 2, 4, 7, and 12 days ahead. The aim of this paper is to investigate how well the ConvLSTM model fairs against a Multivariate Regressor and also to look at whether increasing the length of the sequences of images a model can learn from decreases the prediction error of the model. To establish the effectiveness of the ConvLSTM model, we compare it against the Multivariate Regressor and a Last Frame Regressor model. The results of our work show that the ConvLSTM model outperformed the Multivariate Regressor and Last Frame Regressor models when predicting 1 day ahead, however, when predicting 2, 4, 7, and 12 days ahead, the results of the ConvLSTM and Multivariate Regressor models are quite similar. When comparing sequence lengths, our results show that an increase in sequence length does not necessarily decrease the prediction error of a model.
Docker-Singularity-Containerization
Hi. This project aims to containerize my Twitter-Analysis-using-Pyspark using Docker and Singularity.
Federated-Cloud-for-Water-Resources
This projects requires access to IBM Cloud. The objective of this project is to implement Federated Cloud model for Water Resource Management assuming we are collaborating with the Water Resource Department. Water resource management tool is one of the tools that are essential to the communities as water is one of our vital substances for survival. In this paper we implemented a water resource management tool using IBM cloud services assuming the Water Resource Department chose the path of using the cloud services to digitize all the operations which include three departments, namely, storage department, billing department, and the administrative department where each department is treated like a ”cloudlet”. Application of cloud computing in this operation could be efficient when it comes to costs and offering of agile services. This project can be useful in similar studies and research fields.
House-Prices---Advanced-Regression-Techniques
Used XGBoost to predict houses prices
mcdonaldabdullah
Config files for my GitHub profile.
Scene-Classification-Using-a-Transfer-Learning-Approach
Scene classification means to determine which scene category the contents of an image belongs to. Convolutional Neural Networks, specifically Residual Networks, have proved to be quite usual for the task of image classification. In this paper, we make use of a pre-trained Residual Network to do scene classification. We carry out three experiments in particular; investigate whether data augmentation techniques can improve classification accuracy, whether decreasing the resolution of an image can help prevent object detectors and whether pooling methods like Average Pooling and Max Pooling can improve classification accuracy and also whether they have the ability to preserve spatial information. Results of these experiments showed that data augmentation techniques can improve classification accuracy, decreasing the resolution of an image does not remove object detectors and that average pooling has the ability to improve performance whereas max pooling does not. We also conclude that average pooling has the ability to preserve some spatial information on scene images.
Twitter-Analysis-using-Pyspark
This project data in real time from Twitter tweets about coronavirus and tries to classify the tweets as either offensive or not.
mcdonaldabdullah's Repositories
mcdonaldabdullah/Comparison-of-Predictive-Models-for-Rainfall-Prediction-using-Big-Data-Technologies
Rainfall is a form of precipitation and is responsible for providing most of the freshwater for animals and plants. Machine learning can be used to analyze data trends to develop a model. Deep learning on the other hand focuses more on using images specifically to analyze data. Trying to understand the patterns of rainfall to predict it has proven to be a difficult undertaking, as seen by the various research using machine learning and deep learning for this problem. When implementing a solution to this rainfall prediction problem, a vast amount of computational resources are usually required to execute it. Thus arises a need to properly store and analyze the data to effectively approach the prediction aspect. This paper investigated the comparison of predictive models for rainfall prediction using big data technologies and radar rainfall images. The literature on state-of-the-art prediction models was investigated and compared to survey which models could achieve satisfactory prediction results in combination with big data technologies. The models chosen were Random Forest Regressor and Deep LSTM and were used to predict 1,2, and 3 days ahead using monthly rainfall data. Results from this study showed that the Deep LSTM model performed better than the Random Forest Model for sequence lengths of 4, 8, and 12 when predicting 1, 2, and 3 months ahead.
mcdonaldabdullah/Daily-Rainfall-Prediction-Using-Radar
Reliable daily rainfall predictions can play an important role in a) watershed management, b) disaster management, and c) helping people to plan their day. Numerous factors can have an effect on the patterns of rainfall and therefore it can be difficult to predict. Recent papers studied deep learning for rainfall prediction using various prediction models with an emphasis on short-term predictions (hourly), however, few have looked at daily rainfall prediction using deep learning. This paper discusses a deep learning model, ConvLSTM, for daily rainfall prediction using various sequence lengths of radar images and predicting 1, 2, 4, 7, and 12 days ahead. The aim of this paper is to investigate how well the ConvLSTM model fairs against a Multivariate Regressor and also to look at whether increasing the length of the sequences of images a model can learn from decreases the prediction error of the model. To establish the effectiveness of the ConvLSTM model, we compare it against the Multivariate Regressor and a Last Frame Regressor model. The results of our work show that the ConvLSTM model outperformed the Multivariate Regressor and Last Frame Regressor models when predicting 1 day ahead, however, when predicting 2, 4, 7, and 12 days ahead, the results of the ConvLSTM and Multivariate Regressor models are quite similar. When comparing sequence lengths, our results show that an increase in sequence length does not necessarily decrease the prediction error of a model.
mcdonaldabdullah/Docker-Singularity-Containerization
Hi. This project aims to containerize my Twitter-Analysis-using-Pyspark using Docker and Singularity.
mcdonaldabdullah/Federated-Cloud-for-Water-Resources
This projects requires access to IBM Cloud. The objective of this project is to implement Federated Cloud model for Water Resource Management assuming we are collaborating with the Water Resource Department. Water resource management tool is one of the tools that are essential to the communities as water is one of our vital substances for survival. In this paper we implemented a water resource management tool using IBM cloud services assuming the Water Resource Department chose the path of using the cloud services to digitize all the operations which include three departments, namely, storage department, billing department, and the administrative department where each department is treated like a ”cloudlet”. Application of cloud computing in this operation could be efficient when it comes to costs and offering of agile services. This project can be useful in similar studies and research fields.
mcdonaldabdullah/House-Prices---Advanced-Regression-Techniques
Used XGBoost to predict houses prices
mcdonaldabdullah/mcdonaldabdullah
Config files for my GitHub profile.
mcdonaldabdullah/Scene-Classification-Using-a-Transfer-Learning-Approach
Scene classification means to determine which scene category the contents of an image belongs to. Convolutional Neural Networks, specifically Residual Networks, have proved to be quite usual for the task of image classification. In this paper, we make use of a pre-trained Residual Network to do scene classification. We carry out three experiments in particular; investigate whether data augmentation techniques can improve classification accuracy, whether decreasing the resolution of an image can help prevent object detectors and whether pooling methods like Average Pooling and Max Pooling can improve classification accuracy and also whether they have the ability to preserve spatial information. Results of these experiments showed that data augmentation techniques can improve classification accuracy, decreasing the resolution of an image does not remove object detectors and that average pooling has the ability to improve performance whereas max pooling does not. We also conclude that average pooling has the ability to preserve some spatial information on scene images.
mcdonaldabdullah/Twitter-Analysis-using-Pyspark
This project data in real time from Twitter tweets about coronavirus and tries to classify the tweets as either offensive or not.