Human Activity Recognition (HAR) is a popular area in sensor technology and smart learning algorithms. HAR is the process of recognizing individual or group activities using a set of sensors and an effective learning algorithm. Most of the research on HAR systemsiseitherbasedondatasetscollectedinacontrolledenvironmentorhas minimal human activities associated with them. This paper proposes an efficient and lightweight hybrid deep learning neural network convolutional neural network—long short-term memory (CNN-LSTM) to recognize human activities in daily living using built-in smartphone sensors. A state-of-the-art wearable sensors-based HAR dataset is developed using the inbuilt accelerometer and gyroscope sensors in an unsimulated environment consisting of 20 user data and 8 daily living human activities. Multiple benchmark deep learning models and an intense data pre-processing pipeline are also designed for model validation and to make the dataset usable for classification. The proposed CNN-LSTM achieved an average performance accuracy of 97%, outperforming all the incorporated benchmark models in optimal computational time margin. Further, the training and validation loss was minimal compared to the benchmark models as they lack the combination of spatial and temporal feature handling. The proposed model attained a lower loss percentage of 0.0977% compared to 0.1235% and 0.2569% with LSTM and DNN benchmark models.