Search

Sales forecasting is a vital technology nowadays in the retail industry. With the help of advanced machine learning and deep learning algorithms, business owners can accurately predict the sales of thousands of products and make optimum decisions based on them. In this paper, we proposed a sales forecasting system based on CatBoosting. The algorithm is trained on the Walmart sales dataset, by far the largest dataset in this field. We performed effective feature engineering to boost prediction accuracy and speed. In the experiments, our model outperforms traditional machine learning methods like Linear Regression and SVM, reaching an RMSE of 0. 605. Our method doesn't need as much finetuning as other methods thus improving its generalization ability on other custom datasets, expanding its potential use.

People counting plays an important role in public safety, building automation control and other data analysis like consumer behaviors, passenger management and so on. Among all the public places, the corridor is one of the most common scenes. Therefore, this paper proposes a people counting system which is applied in the corridor. The proposed system counts the amount of the passing pedestrians as well as detects the direction of each pedestrian. A mirror reflection device is designed to achieve a double-plane scanning by a single lidar, which doesn't only double the scanning data and make use of the wasted scanning angle, but also realize the direction detection by judging which scanning plane the pedestrian has passed first. A novel shape descriptor is used in the sliding window algorithm, and a support vector machine is trained as a classifier. In the proposed algorithm, head-shoulder feature in the pedestrian point cloud is detected to get the point cloud of every pedestrian. In the experiment, actual pedestrian data are collected to confirm the proposed system. The proposed method shows a better performance than the comparison methods.

Handover (HO) is one of the key procedures, which ensures user equipment (UE) moving smoothly in long term evolution (LTE) systems. Generally, hard handover has a high outage probability and is relatively simple, but when the velocity of UE is high, the procedure may be unreliable, especially for VoIP service. In the special scenario high-speed railway, network deployment is known and an optimized relay algorithm could be applied according to information already known, such as cell deployment and direction of UE running. Simulation results show that algorithm proposed performs better than the traditional hard handover in high-speed railway networks.

One of the main goals of LTE is to provide seamless access to voice and multimedia services with strict delay requirements, which is achieved by supporting effective handover algorithm from source cell to target cell. In this paper we first give an overview of the LTE architecture and the LTE intra-frequency handover procedure. Secondly, we analysis the performance of A3 handover algorithm based on A3 Event from the handover failure rate. Then, we propose a new handover algorithm which can reduce the pingpang handover compared to A3 algorithm.