Data Scientist

• Digital Insights and Innovation Consultant for Northern Province Provide technical and scientific knowledge and advice on the progress review dashboard of all ongoing projects and the action procedures to be followed for planning future projects.
• Clustering algorithms for spatial data Scikit-learn clustering package was used to define the meaningful clusters for spatial data. The characteristics of the clustering methods were defined for three main use cases: pre-defining the no. of clusters, initiate the cluster centroids, and imposing restrictions on clusters.
• Technologies: Python

• OCR and Computer Vision based model for scanned documents Scanned pdf documents were used as the input data to provide machine encoded text for several NLP tasks. After converting the pdf as images tesseract OCR and OpenCV-python was used to extract the text from the images. Several mitigation methods for different orientation and background noise were used to increase the OCR output accuracy.
• Document classification model based on Artificial Neural Network Convolutional Neural Networks (CNN) was used to identify the structured layout of the scanned documents. Transfer learning was used to decrease the training time for the deep convolutional neural network models by reusing the model weights from pre-trained models.
• Semantic Similarity and Summarization (NLP) Cosine similarity was used to measure the similarity between two paragraphs. Transformer models were used to encode sentences to get their embedding meaning of words. Then the paragraphs that contain similar semantics were summarized by using abstract based summarization. Machine Learning for Sales Forecasting, Machine Learning approaches were used to forecast the next 12 months sales in beverage industry by considering both external and internal factors. Macro-economic factors were considered as the external factors, and promotional events and competitor actions were considered as the internal factors. Model was built to do the forecast up to outlets level.
• Technologies: Python

Project:

• Machine learning for inflation forecasting Elastic Net, Random Forest, XGBoost, and LSTM models were fitted and selected the best model considering the MSE (Mean Square Error). Feature importance analysis was conduct to identify the important features to forecast the inflation based on the best model.
• Technologies: Python, Excel