SHREYAS SUDARSAN

BITCOIN PRICE PREDICTION

• Used a bitcoin price dataset containing variables such as Lowest price, Highest price, Close price, Open price and Volume
• Created a Future price data variable by shifting the Closed price data down by 3 data points
• Created a Rise/Fall data field by comparing Future and Closed price variables
• Preprocessed the data using MinMaxScaler and created Buy and Sell lists
• Used a multi-layered LSTM neural network along with Batch Normalization and Dropout layers
• The model predicted if the Bitcoin price would rise/fall for a given data point with a test accuracy of 92%
• Software and packages used: Python, Keras, Pandas, Numpy

MANGO LEAVES DISEASE DETECTION USING CNNs

• Main objective of the project was to build a CNN model that could predict the presence of disease in a Mango Leaf image.
• Used a Kaggle dataset of mango leaf images that were healthy and diseased.
• Used an Image data generator to augment the dataset.
• Built a multi-layer Deep Neural Network using Convolutional, MaxPooling2d, and Dense layers, achieving an accuracy of 97% on the validation dataset.
• Presented a conference paper on this project at the International Conference on Innovative Computing and Communications 2021, and published the same as part of the Springer AISC series.
• Softwares and packages used: Python, Keras, Tensorflow, Numpy

NON INVASIVE GLUCOSE MONITORING SYSTEM

• Developed a non-invasive technique to detect the blood glucose level of a patient
• We fabricated a dielectric spectroscopy sensor which helped in calculating the blood glucose level without requiring the need for a blood sample
• We also built an RNN model which forecasted time and helped in approximately predicting the glucose levels using past patient data.
• Researched different glucose measurement techniques
• Researched dielectric spectroscopy, Kalman smoothing filters
• Sensor data conversion and RNN model for approximately forecasting future glucose levels
• Software and packages used: Python, Keras, Tensorflow